Theory of Constraints Production Drum Buffer Rope

3/12/2014 Theory of Constraints Production Drum Buffer Rope

http://www.dbrmfg.co.nz/Production%20DBR.htm 1/59

A Guide to Implementing the Theory of Constraints (TOC)

PowerPoints Preface Introduction Contents Next Step Advanced

Bottom Line Production Supply Chain Tool Box Strategy Projects & More ... Healthcare

Drum BufferRope

ImplementationDetails

Batch Issues Quality/TQM II Alignment Time

A Motor For Production

Drum-buffer-rope is the Theory of Constraints production application. It is named after the 3essential elements of the solution; the drum or constraint or weakest link, the buffer or materialrelease duration, and the rope or release timing. The aim of the solution is to protect the weakestlink in the system, and therefore the system as a whole, against process dependency and variationand thus maximize the systems overall effectiveness. The outcome is a robust and dependableprocess that will allow us to produce more, with less inventory, less rework/defects, and better on-time delivery always.

Drum-buffer-rope however is really just one part of a two part act. We need both parts to make areally good show. If drum-buffer-rope is the motor for production, then buffer management is themonitor. Buffer management is the second part of this two part act. We use buffer management toguide the way in which we tune the motor for peak performance.

In the older notion of planning and control, the first part; drum-buffer-rope, is the planning stage of theapproach essentially the overall agreement on how we operate the system. The second part,buffer management, is the control system that allows us to keep a running check on the systemeffectiveness. However, I want to reserve the word planning and the word control for quitespecific and established functions within the solution, functions that we will investigate further on thispage.

I want to propose that we step out a level and instead use the terms configuration andmonitoring. Using this terminology the configuration is drum-buffer-rope and the monitoring isbuffer management. Lets draw this;



The way that we configure the solution, the way that we configure the; drum, the buffer, and the rope,will determine the characteristics and the behavior of the system as a whole. Buffer managementallows us to monitor that behavior. The use of the terms configuration and monitoring will allow amore critical distinction to be developed once we introduce the concepts of planning and control. This, I hope, will also clarify some of the confusion that may exist over the dual role of buffermanagement.

Keep this model in mind as we will return to it. Now, however, we must return to our plan of attackand work through the development of the solution.

Interested? Then lets go.

Our Plan Of Attack

On the measurements page we introduced the concept of our rules of engagement which is todefine; the system, the goal, the necessary conditions, the fundamental measurements, and the roleof the constraints. Then on the process of change page we introduced the concept of our plan ofattack the 5 focusing steps that allow us to define the role of the constraints. Lets remindourselves once again of the 5 focusing steps for determining the process of change;

(1) Identify the systems constraints.

(2) Decide how to Exploit the systems constraints.

(3) Subordinate everything else to the above decisions.




(4) Elevate the systems constraints.

(5) If in the previous steps a constraint has been broken Go back to step 1, but do not allow

inertia to cause a system constraint. In other words; Dont Stop.

Lets also return to our simple system model which we have so far used in much more general termsand apply it to drum-buffer-rope. As you will recall it has 4 sections, or departments or whatever youwould like to call them; a beginning, a middle, a near-the-end, and an end.

Our system has to interact with the outside world, so lets draw in an input and an output. Rawmaterial flows in and finished product flows out. In a for-profit situation sales flow in and expensesflow out, the difference profit, is captured by the system. We showed these flows previously in thesection on measurements.

Now we are ready for the next stage, the first step in the 5 step focusing method identify theconstraint.

Identify The Systems Constraints

In fact we know where the constraint is in our simple system presented here based upon thediscussion in the earlier section on measurements. Its located near the end of the process. Thisisnt at all an unusual place to find a constraint. Think about it for a moment. If the constraint waslocated near the beginning, then all the downstream steps would always be waiting for work. In that

situation management would most probably go about purchasing further capacity until they move the



situation management would most probably go about purchasing further capacity until they move theconstraint further down the process and then bury it in work-in-process so that it is no longer visible.

Lets draw the constraint in.

As we know from the previous section on production, the constraint, the slowest step, beats out therate at which the whole process can work at. Therefore it becomes the drum of drum-buffer-rope.

Of course we forgot something work-in-process. If our model system is to be anything like ourown reality, then it is probably full to the gills of work-in-process. We had better add this to ourmodel as well.

Work-in-process of course serves a useful purpose in such a system; it decouples each stage fromthe stages before and after. If you dont know what to protect, then you might as well protecteverything. However chances are that, even with all that protection, the work that was required at thetime wasnt the work that was waiting in the pile of work-in-process. And of course it means that thetime required for any job to traverse the system is much longer than necessary. In any case we dontneed all of that work-in-process anymore if we are going to use drum-buffer-rope.

So we have completed Step 1 identify the constraint. The next step, step two, is to decide how toexploit the constraint.

Exploit The Systems Constraints

To make sure that the constraint works as well as possible on the task of producing or creatingthroughput for the system we must ensure that we exploit it fully essentially we are leveraging the

system against the full capacity of the constraint. This means not only making sure that it is fully



system against the full capacity of the constraint. This means not only making sure that it is fullyutilized, but also making sure that the utilization is fully profitable. If you remember back to the P & Qproblem or the airline analogy, is quite possible to have everything utilized but not make as muchprofit as is possible.

If we increase the output of the constraint, then the output of the system as a whole will increasealso. One of the most effective tactics for exploiting the constraint, once identified, and improving itsoutput is to write a detailed schedule for that particular resource and that particular resource alone and then to adhere to that schedule. This is the plan in this context. Our day-to-day planning fallsout as a consequence of the decisions that we make while configuring the implementation. Letsadd this to our model.

We now have a local plan for just one point, the most important point, the drum. If, at the same time,we hold the input constant then the additional output from continued exploitation must come fromwork-in-process already in the system. As a consequence work-in-process and hence lead timesmust go down. In effect we begin to drain the system. Lets show that.



Lets be clear however, work-in-process does not have to decrease under drum-buffer-rope, butusually there are sound reasons for doing so reduced lead times, increased quality, andincreased throughput. We will investigate all of these sometime later under the heading of the roleof inventory. The primary objective of Theory of Constraints however is always to move the systemtowards the goal, and usually that means increasing throughput first. Inventory reduction issecondary and often a consequence of increasing throughput.

If we continue to operate in this fashion we can reduce work-in-process considerably. Lets showthis before introducing some further drum-buffer-rope concepts.

In fact we have completed the second step; we have decided how to exploit the constraint. Weused a simple example of writing a schedule, there are many more ways to exploit a constraint andsome of these are mentioned in the next page on implementation details. However, we need tomove on to the third step, subordination of the non-constraint resources.

Subordination Protect The Systems Constraints

Sometimes using the word protect makes it easier to understand this step than using the correctterm which is subordinate. In fact, we subordinate the non-constraint resources in order to protect

the constraint and the system as a whole. Lets examine this is a little more detail.



In the process of change page we described subordination as avoiding deviation from our plan, andthe plan in this case is our constraint exploitation schedule in the previous step. We describeddeviation from plan as (2);

(1) Not doing what is supposed to be done.

(2) Doing what is not supposed to be done.

We can therefore describe subordination as;

(1) Doing what is supposed to be done.

(2) Not doing what is not supposed to be done.

By doing what is supposed to be done in accordance with our plan we protect the constraint and thesystems as a whole. Moreover, by not doing what is not supposed to be done in accordance withour plan we also protect the constraint and the system as a whole. Lets examine this with oursimple model.

As we use up our supply of excess work-in-process, it is likely that the constraints will begin tostarve from time to time. Work will not arrive in sufficient time for it to enter the constraint onschedule. We need to replace our local safety everywhere (our excess work-in-process) with someglobal safety right where it is needed, in front of the constraint. We need to buffer the constraint.

We need to do what is supposed to be done in order to protect the constraint from shortages.

In fact we would normally have made our buffering decisions before we even began and thereforereduced our work-in-process and lead time in line with these pre-determined targets.

An Initial Buffer Sizing Rule

Lets assume for a moment then that the lead time allowed for work to travel from the start of theprocess to the start of the constraint was 18 days prior to the implementation. Well, in fact, it couldbe 18 hours for electronics or the paper work in an insurance claim, or it might be 18 weeks forheavy engineering. But lets use days in this example. The rule of thumb to apply is to halve theexisting lead time (3). Therefore the new lead time becomes 9 days. If halving the lead time soundshorrendously short, it is not. Most of the time the current work-in-process is sitting in queues doingnothing. You can easily check this for yourself got out and tag some work with a flag or a balloonor a bright color and then watch it. It will sit. This 9 day period becomes our buffer length.

To this 9 day buffer we apply a second rule of thumb and divide the buffer into zones of one thirdeach (4). We expect most work to be completed in the first 2 thirds and be waiting in front of theconstraint for the last third of the buffer time. Thus we expect our work to take about 6 days ofprocessing (and waiting-in-process) and 3 days of sitting in front of the drum.

If 3 days sitting in front of the constraint sounds terrible, then remember that prior to theimplementation, the system allowed work to sit for at least another 9 days. Nine plus 3 is 12 days

sitting. Which would you rather have 12 days or 3 days? More importantly, which would yourcustomer prefer?



customer prefer?

We now can protect our system constraint by ensuring that there is always work for it to do. Thus weensure its effective exploitation and with much less total material or lead time than before.

Lets add the buffer to our diagram.

Lets make sure we are clear about the definition of the buffer. For all practical purposes the TIMEBUFFER is the time interval by which we predate the release of work, relative to the date at whichthe corresponding constraints consumption is scheduled (5).

Please be careful, on the diagram above we have drawn units of time the zones and the buffer as space on our diagram. Dont let this confuse you. The zones equate to time allocated in theplant to protecting an operation whose position and function is critical to the timeliness and output ofthe whole process. The zones do not equate to the position of work in the plant. In fact we will returnto this shortly and try and draw the diagram more realistically to represent time.

Why is this whole period from material release to the constraint considered as the buffer? Schragenheim and Dettmer consider that this is one of two unique aspects of buffering in Theory ofConstraints. The reason buffers are defined as the whole lead time and not just the safety portionis that in most manufacturing environments there is a huge difference between the sum of the netprocessing times and the total lead time. When we review the net processing time of mostproducts, we find it takes between several minutes and an hour per unit. But the lead time may beseveral weeks, and even in the best environments several days. Consequently, each unit of productwaits for attention somewhere on the shop floor for a much longer time than it actually takes to workon it. So it makes sense not to isolate the net processing time, but to treat the whole lead time asa buffer the time the shop floor needs to handle all the orders it must process (6).

The other unique point is that buffers are, as we have mentioned, measured in time. Firms in non-

drum-buffer-rope settings consider a buffer to be a measure of physical stock; 6 jobs, or 6 orders, or10 batches, or 4000 pieces, or whatever. In drum-buffer-rope a constraint buffer is a measure of



10 batches, or 4000 pieces, or whatever. In drum-buffer-rope a constraint buffer is a measure oftime; hours or days of work at the constraint rate located between the gating operation (materialrelease) and the constraint. In fact, there are two ways to look at a buffer, either from theperspective of a single job, or from the perspective of the system as a whole. Lets consider this fora moment.

Lets assume for the sake of simplicity that all of our jobs are of equal length. Lets assume then thateach one takes 1 day of constraint time. In this case each job has a 9 day buffer to the constraint. That is, it is released 9 days prior to its scheduled date on the constraint. This is the perspective ofa single job. The constraint, looking back, will see 9 one-day jobs at various stages in the process;this is the perspective of the system as a whole.

What then, all else being equal, if all of our jobs now take half a day on the constraint? Each jobsees a 9 day buffer, the constraint looking back will see 18 half-day jobs at various stages in theprocess, but the aggregate load is still 9 days, this is the perspective of the system as a whole.

Lets do this one more time. Each job now takes quarter of a day on the constraint. Each job stillsees a 9 day buffer, the constraint looking back will see 36 quarter-day jobs at various stages in theprocess, but the aggregate load is still 9 days from the perspective of the system as a whole. It istime that is the measure of the buffer.

Why Is The Constraint Buffer Size & Activity Determined By Time?

Lets labor this point for a moment because it is so important. Measuring a constraint buffer in unitsof time is unique to drum-buffer-rope because acknowledgement of the existence of a singularconstraint within a process is unique to drum-buffer-rope. We can apply this to both the constraintbuffer size and the constraint buffer activity.

Lets look at constraint buffer activity first.

By considering only one station, or step, or procedure, we need only to know one set of averagetimes for that place or action for all of the different types of material units that pass through it. Wecould look at this as follows;

At a manufacturing constraint an hour is an hour but the number of units may differ

The number of physical units may differ because different types of material using the sameconstraint may use different amounts of constraint time. In fact, even the same type of material willdisplay some variability unless the constraint is a totally automated procedure but these will largelyaverage out.

How about constraint buffer size then?

The unique perspective brought about by the designation of a singular constraint allows us to definethe length of the buffer in time also. Essentially the buffer is sized and sees the duration from thegating operation to the constraint due date. Moreover the buffer sees committed demand workthat has already been released to the system. Constraint buffers, divergent/convergent control point

buffers, assembly buffers, and shipping buffers are all of the same basic nature.



Maybe it is much simpler to say that;

We protect time (due date) with a time buffer

There is, however, one other buffer type that we are likely to come across in manufacturing a stockbuffer. There are two places that these occur at in manufacturing; they are at raw material/inwardsgoods in all process environments and at finished goods in a make-to-stock environment. Theseare actually supply chain buffers; they represent the two places that the supply chain must interactwith processing before the beginning of the process and after the completion of the process. Weneed to ensure that we always have an adequate supply of raw material prior to the process to meetconsumption and we need to ensure that we always have an adequate supply of finished goodspost-production to meet demand. We will examine these types of buffers later on this page. Theyare also examined in more detail on the supply chain pages especially the replenishment page. However, lets confine ourselves at the moment to constraint buffers. We need to labor the issuethat the constraint buffer is a measure of time. Lets do that.

A Journey Through Time

Many, many, people say that they do understand the definition of a drum buffer or of a constraint

buffer when the evidence is that they do not. Too often our prior experience causes us to think ofbuffers in terms of physical stock, and too often we consider zone 1 as the buffer. Lets see.

The buffer is the whole of the duration of the part of the system that the buffer protects.



Did I overstress the point? I dont think so. Check here for more discussion on continual mis-understanding of buffers in drum-bufffer-rope.

In part, this is due to our prior manufacturing experience with MRP II systems and push productionwhich tends to blind-side our interpretation (see the sections on Buffer The Constraint and LocalSafety Argument in the next page Implementation Details for further development of thisaspect.) In part, the problem also lies in the way we try to draw time as space on our simplediagrammatic representations. The only way to draw time is to draw a sequence of diagrams. Lets do that.

We will follow a slice of work ones days worth through the process to the drum. We will use our9 day buffer as we derived above, so this slice of work is the drums work for one day 10 days outfrom the scheduled processing date. There are 5 products (units, jobs, batches, whatever) in ourslice. The products are lilac, red, green, blue, and orange. The time interval, for the sake ofclarity in this example, is course days rather than finer divisions of hours or less that we mightexpect to find in reality.

Imagine that within the departments (beginning and middle) of our generic process we have thetools of our particular trade; be they desks in a paper trail, admission or beds or clinical units in ahospital, or work centers in a manufacturing system. The 5 products could be at any time waiting ormoving between jobs or being worked upon. The resolution of this detail isnt important to us here.

Probably in the day before the release date the planner knows what will be released. The plannermight even have the orders cut and waiting but unreleased (and hopefully unknown to the floor toavoid people working ahead of time). Lets draw this.



The orders may exist on a plan but they are not yet released. We draw the units outside of thesystem even if they currently have no physical presence other than paper work or an entry on ascheduling system.

We have also drawn a timeline in. It is colored according the buffer zones. Zone 3 is the greenzone, zone 2 is the orange or yellow zone, and zone 1 is the red zone.

On the first day of the schedule all the products are released (as scheduled) and are in zone 3 of ourtime buffer. Their physical location at the end of the day is as follows.

Lilac might be small batch or a simple process that is completed quickly, it moves forward further(and maybe faster) than the rest.

After another day we are at day 2 and still within buffer zone 3 the process looks like this.



We can see that red has moved quite quickly relative to the others and blue hasnt moved at all. How does this happen? Different jobs travel through different routings, and have different wait times(because of other jobs in front of them) and different processing times (either because of differentbatch size or different work). And of course sometimes things dont always go as planned; we havebreak-downs, people are absent, and stuff happens.

By day 4, one day into buffer zone 2 we see the work has evolved as follows.

Blue still hasnt moved, however, the others are progressing well.

The next day, day 5 (buffer zone 2), the work looks like this.



The green and red jobs are complete, the lilac and orange jobs are progressing well and blue ismoving forward at last.

At the end of day 6 the last day of buffer zone 2 we see the following.

Three of the five jobs are completed by the end of buffer zone 2, and two are lagging behind.Because the end of day 6 is the starting of day 7 and the first day of zone 1 (the red zone) we have abuffer penetration. Two jobs that ought to be finished by now have not been finished. They must belocated and appraised to ensure that they will reach the constraint in the remaining time.

Lets go out to the end of day 7 and see what happens.



Now 4 jobs have been completed. The lilac job was completed sometime on day 7. The blue jobwill have been located and checked to see that it will meet the schedule and be available at thedrum by the end of day 9 the last day of zone 1 or preferably sooner. We would have preferredthat most jobs were completed by the end of day 6, but sometimes stuff happens and not all jobsare complete at that time. The blue job might require some assistance to ensure its completion.

Lets now look at the situation at day 8.

Phew! We find at the end of day 8 with just one day to spare that all of the jobs are completedand waiting to be processed on the constraint according to the schedule on day 10. Zone 1 thered zone of the buffer was penetrated by as much as 2 days by the blue job and as much as 1 dayby the lilac job. However the drum is now fully protected and the drum schedule will not be

compromised, our exploitation strategies are fully protected by adequate subordination of the otherresources.



So, lets reiterate once more; buffer zones equate to the time allocated in the plant to protectingoperations whose position and function are critical to the timeliness and output of the wholeprocess, buffer zones do not represent the physical location of work in the plant.

We know how to protect the constraint, now lets see how to protect everything else.

Subordination Protect Everything Else

So, we know now how to protect the constraint using a buffer, we therefore know how to do the firstpart of subordination doing what we are supposed to do in order to protect the constraint. Nowwe need to examine the second part of subordination not doing what we are not supposed to do.

First, lets repeat the diagram that we first drew prior to our journey along the buffer.

The drum, gating operation, and shipping are all stable. That is they are now all operating at thesame rate, the drum beat. If we were successful in exploiting the constraint and increasing theconstraint rate and output, and demand increased as a consequence of the reduced lead time, thenat this stage the input rate must be also be increased to match the drum beat so that work-in-process and the buffer remains stable.

In order to maintain stability in this system, the rate at which the gating operation allows theadmission of new work to the system must be the same as the rate of consumption at the drum. What would happen then if our constraint breaks down for a short period? It has no spare capacity,so we cant speed it up (allow it to work longer) to catch up to the work again. If we were to continueto admit work as though nothing had happened then work-in-process would increase a little. Probably not enough to notice, but over a couple of different instances it would begin to build up. Thus we need to make sure that we admit new work into the system at the same rate as the

constraint is consuming it. The constraint as we know is the drum of our system, beating out therate at which the whole system works, including the gating operation. The rate is communicated tothe gating, or first, operation by the rope. We need to add this to our diagram.



the gating, or first, operation by the rope. We need to add this to our diagram.

If you like, the schedule of the gating operation is the schedule of the drum off-set by a rope length oftime. The rope length is the same as the buffer duration; the gating rate is the same as the drumrate. Tying the rope between the drum and the gating operation ensures that excess work can notbe admitted, or that normal work can not be admitted too soon. This is part of not doing what isnot supposed to be done in order to protect the system from excess work-in-process. Excess

work in process results in longer than necessary lead times and poorer quality. Ultimately excesswork-in-process also impacts upon the throughput of the constraint.

Another way of looking at the rope is to consider it as a real time feedback loop between the drumand the gating operation.

Although the constraint can not recover from down-time, hence the need to exploit and protect it, thenon-constraints can recover from down-time. Generally the non-constraint parts of our system dontwork at the same rate as the whole system at least over short periods of time. The non-constraints have sprint capacity. They can and do process more work when necessary to catch-upafter a bump in the system by operating at normal rates for longer periods. They can also processless work when not needed by operating at normal rates for shorter periods.

We might consider the increased duration of non-constraint processing (in order to catch up) as thedoing what is necessary part of subordination, and the reduced duration of non-constraintprocessing (to avoid over-production) as the not doing what is not necessary part ofsubordination. It is critically important that we do this.

In the Toyota production system, kanban perform both of these functions. In drum-buffer-rope this isperformed by the roadrunner concept. When we have work to do, we do it. When we dont havework to do, we dont do it. We saw this in the form of the traffic light analogy earlier in the page forprocess of change. Non-constraints should never slow down, they should either be fully-on or fully-off (maybe that should be normally-on or normally-off). Either creating throughput or protectingthroughput. We will revisit this theme on the next page; implementation details.

We have seen how we now have 9 days of work in process in our example down from the initial 18days. There are 6 days of work in process in zones 3 and 2 and three days in zone 1. But we canthink about it in another way. By halving the work-in-process we have removed 3/6ths of the workfrom the system. We have another 1/6th sitting in front of the constraint, so in effect we have just2/6ths or 1/3rd of our previous work-in-process actually on the floor being actively worked on orsitting in queues. Imagine your process at it stands today but with just 1/3rd of the work activelybeing worked on by the non-constraints or waiting-in-process; wouldnt things really begin to flow inthat situation?

Nevertheless, in order for that material to flow, it is critically important to protect sprint capacity byproper subordination. Sprint capacity interacts with overall buffer size and hence manufacturinglead time. We will investigate sprint capacity more in the next section. Protecting sprint capacitymeans that we never admit work into the process just to keep people busy never.

Of course you will have noticed that we have so far used the departmentalized version of our system

model. To some extent this was intentional because when you first approach a drum-buffer-ropeimplementation, the process will be departmentalized. However, what we would like to see after ashort while is a better appreciation of the system as a whole.



short while is a better appreciation of the system as a whole.

Conceptually it should look more like this;

Until now we have ignored that part of the process after the constraint usually from the constraint toshipping in make-to-order, or to the warehouse in make-to-stock. This part of the process is tied tothe shipping date by a second rope most often referred to as the shipping rope. It is via this ropethat the drum is tied to the market demand. So now we can be sure that just enough new material isallowed to enter the system to protect and satisfy the drum consumption, which in turn supplies themarket demand. We dont admit work for which there is no demand. Therefore, we have indeedsubordinated everything else to the constraint.

For the shipping buffer, we again apply the same rules of thumb as we used for the constraintbuffer. Lets assume, for the sake of the ease of the mathematics, that the process downstreamfrom the constraint to shipping currently takes 6 days. We halve that to give us a new lead time of 3days. And then we divide that into buffer zones of thirds and expect almost all work to be completedafter 2 days and either waiting for shipment or already shipped 1 day prior to the shipping date.

This is what we arrive at;



Now our shipping date, or delivery date, is protected as equally well as the constraint is. We shouldtherefore expect very good on-time delivery or due date performance. This of course is especiallyimportant for make-to-order firms and is most often a definite competitive advantage.

Thus our original 24 day process becomes a quoted delivery time of 12 days. The system shouldbe able to produce more because we will have made every effort to exploit the constraint, and thenon-constraints only work on material that is required to support the constraint schedule.

So, to summarize, subordination is the instruction to the non-constraints. It has two maincomponents.

Firstly; in order to protect the system as a whole we must not starve the constraint we must notunderload the system. This will ensure maximal output as per the exploitation strategy of theconstraint. Of course we could make the buffer quite large and never have to worry about starvingthe constraint, but that is where most systems are today (and they still starve the constraint). So weneed to do something else as well. Secondly; in order to protect the system as a whole we must notflood the non-constraints we must not overload the system. This will ensure adequate sprintcapacity to ensure maximal output as per the exploitation strategy of the constraint and high duedate performance. It will also ensure a much reduced lead time.

Thus we have covered all three aspects; the drum, the buffer, and the rope. We have also coveredthe first three steps of the 5 focusing steps; identify, exploit, and subordinate.

If we have fully exploited the leverage point, and subordinated everything else, then the next thing todo is to elevate the constraint. But first, lets examine an alternative initial buffer sizing rule.

An Alternative Initial Buffer Sizing Rule

So far we determined our initial buffer sizes by taking 50% of the existing lead time over the part of

the system that we wished to protect. There is another lesser-known rule that we can also use. Wecan take 3 times the back-to-back time for a job to transverse that part of the system that we wish toprotect.



From time to time jobs are expedited for a number of reasons, therefore people will know fromdirect experience, or will have good intuition, for the back-to-back time and from that a bufferduration can be obtained.

Elevate The Systems Constraints

Elevation may require that some additional investment be made to purchase additional capacitythat will produce additional throughput, preferably at more than pro-rata. Remember we are trying todecouple throughput from operating expense thereby driving productivity and profitability up;

We also know that elevation is most often the place where reductionist/local optima proponentsstart, however elevation is the place where systemic/global optimum proponents get to last. It costsyou less doing it this way, of course you have to think but then its just common sense and youmake more money or more output. If we dont decouple throughput in for-profit organizations andoutput in not-for-profit organizations from operating expense and investment then we simply are notdoing a very good job.

If A Constraint Has Been Broken, Go Back

If at any time a constraint is broken then we must look for the next constraint. In fact we should knowfrom buffer management where it will be before we even get there. However, many times an initialphysical constraint is broken and a policy issue takes its place. Goldratts admonishment not to letinertia become a system constraint is a plea to look at which policies block us from moving forwardeven further. Really this is a plea to Dont Stop!

Some Definitional Subtleties

There are a few traps for those of us who are new players. Some of the definitions have changedover time. In this case to be forewarned is to be forearmed, lets do that. Here we have used theterm drum to describe the entity that determines the rate at which the whole system works be thatan internal constraint, or as we will see, an external market demand. However the term drum is

also used to describe the drum schedule, in fact some insist that the drum is the schedule. Clearlywhen the constraint is in the market this definition makes more sense. Both are in use, some wouldargue that these different definitions are simply different manifestations of the same concept. If weaccept this, then that ought to keep everyone happy.

Likewise, the rope has been used here to describe the off-set between the drum and the gatingprocess; however, it too is subject to a more restrictive definition of the gating schedule. Once

again these are different manifestations of the same concept. They are not mission critical.

Buffer Management Make-To-Order



Buffer Management Make-To-Order

We have so far examined the development of the drum-buffer-rope solution our motor forproduction and we have done that within the framework of our 5 step focusing process. We alsopresented a model for the configuration and monitoring of this solution, to this we have added alocal planning aspect; our schedule for the exploitation of the drum. Lets repeat the model here andnote that we are looking at the specific case of make-to-order.

We now need to develop the monitoring part of his model; we need to address buffer management. We now know what buffers are and we know their purpose, however we still need to know betterhow to interpret and utilize the information that they can provide. And in order to do that I believethat we must subdivide their impact into two distinct functions. They are as follows;

(1) Local Control; the day-to-day exception reporting that indicates when there may be apotential due date violation.

(2) Global Feedback; longer term trend-reporting that suggests a particular buffer needs to beresized to be fully effective.

Buffer management is crucial; it filters important signals from the day-to-day noise of the systemthereby alerting us to potential problems before they become real problems and it provides a self-

diagnosis that neither too much and nor too little protection is made available for each case. Theself-diagnosis feeds back into our configuration and guides improvements in the overall dynamicsof the implementation.



Lets modify our model to incorporate this.

Thus we still have planning and control, but it is local and within the context of the overall design ofthe implementation. Schragenheim & Dettmer have an important definition of control, lets repeat ithere (7);

A reactive mechanism that handles uncertainty by monitoring information that indicates athreatening situation and taking appropriate corrective action before the threat is realized.

The subdivision of buffer management into local control and global feedback will, I hope, make iteasier to understand this important concept.

Lets now investigate the various stages of local control and then global feedback in a make-to-order environment.

Local Control Buffer Status

We release work a rope length ahead of the due date at the point that the buffer protects. In mostcases that is more than sufficient to insure that the work arrives in good time. But as we know

sometimes stuff happens. We need some local control to ensure that when stuff does happen weknow the correct priority to return the system to one of stability. We need to know the status of thecurrent make-to-order work orders that are already released to the system to be worked upon.



Schragenheim defines the buffer status as follows (8);

Buffer Status = (Buffer Duration Remaining Duration) / Buffer Duration

The buffer status is synonymous with buffer consumption.

Lets look at an example. Here we have the buffer status for 6 jobs due in the next 5 days.

Jobs 1, 3, & 5 are completed as of today. For jobs 3 & 5 that means they were completed in goodtime, their buffer status is less than 66% at completion. Job 1, although now complete, must havebeen problematic a day or so ago as its buffer status had reached 78%, however it is nowcompleted and no longer a problem.

Of the current jobs; job 6 with a buffer status of 44% is no problem, we should leave it alone. However jobs 2 & 4 both have a buffer status of 66% or more. They have both begun to penetratezone 1 of the buffer, the red zone. Of the two, which is the greater priority? Clearly it is job 2; this isthe one with the greater buffer status value and is the one that we should first concentrate ourattention on.

Thus the buffer status drives the work order status once the job has been released to the system. We therefore have local control by exception in order to meet our global objectives. Buffer status isviewed from the perspective of a single job.

Thats Nice But We Have Short Ropes & We Have Long Ropes

What then, where we have short ropes and long ropes, surely that is more complicated? Again thebuffer status allows the direct comparison of work orders that have differing buffer lengths within thesame process. Lets have a look.



We have jobs in the system which have 3 different rope lengths 9, 6, and 3 days. The shorter jobsgenerally go through far fewer processes than the longer jobs. Some short jobs, job 4 for instance,will be released later and finish earlier than longer jobs, such as jobs 5 or 6. Nevertheless, we knowwhich jobs need attention; those that have used up the first two zones and penetrated zone 1 of thebuffer, the red zone. Those in zone 2, the yellow or orange zone, may require a watching brief butwe leave them alone least we begin to tamper with the process and we wouldnt want to do that.

Dont confuse short jobs and long jobs with rush and standard orders. That particularprioritization takes place in the order queue and not in the manufacturing lead time.

Local Control Work Order Status

In a make-to-order system where there is a commitment to meet a due date, then;

Current Work Order Status = Buffer Status

It may seem redundant to state this, however, it will become clearer when we deal with make-to-stock. In a make-to-stock environment the stock order status can vary over time due to the naturalvariability of the process itself and due to changing market demand as well. This additional factor isabsent from the make-to-order environment, here we have a commitment and we must meet thatcommitment.

Local Control New Work Order Release Priority

Ever been in the situation where you have a nice workable schedule all the critical areas havebeen satisfied (all the squeaky wheels are oiled), and then someone announces you know thematerial that they said would be on the truck this morning, well its not! What are you going to do? This job is down for the gating operation this morning, you dont have it, and its the Friday before a



This job is down for the gating operation this morning, you dont have it, and its the Friday before along weekend. Sound improbable? Not at all.

Officially you can wait. You can wait until up to 50% of the buffer to the next stage has beenconsumed. Until then the work can still be released, after that it must be rescheduled.

In fact the same logic can be applied when back-scheduling from the shipping date. Sometimescommitment to different customers and/or varying rope lengths will result in a conflict where morethan one job simultaneously requires the constrained resource. One solution is to start some workon the constraint even earlier than required. If a conflict still exists then some work may have to startlater than we would like, in effect the late start pushes that job out so that it begins to consume partof its shipping buffer. Again, so long as not more than 50% of the shipping buffer is consumed thenthe job can proceed, otherwise the commitment to the client must be renegotiated the due datemust be pushed out further (9).

Stepping back one step further, the same logic applies in-turn when back-scheduling from the drumdate to material release, now for a different reason than material lateness; we can apply the samerules as before. In fact, this is really local planning rather than local control, but it is easier tounderstand now rather than earlier.

Local Control/Global Feedback Work Order Zone 1 Buffer Penetration

We want to insulate our customers from the variability and dependency within our system. We cando this in two ways; (a) proper subordination so that non-constraints have adequate sprint capacity,and (b) buffering so that even when sprint capacity is exhausted for what ever reason (stuffsometimes happens) we still have some safety time up our sleeves. Of course, we have rolled-upour local safety everywhere into a few discrete and critical places where it will be of maximal benefit.

Work order status our buffer status tells us in real time when to facilitate an order, and in whatpriority we should do so if more than one work order requires facilitation at the same time. It is local,immediate, and pre-emptive. A buffer status that exceeds 66% has consumed more than 2/3rds ofits buffer capacity, colloquially we say that it has penetrated the red zone.

Until the order is complete and at the buffer origin we dont know to what depth the red zone hasbeen penetrated. However, once the order is complete and the degree of penetration is known wecan call this degree of penetration a buffer hole. Lets draw this.



Buffer holes are our measure of system stability. They are global, aggregate, and historic. Trendsin buffer hole behavior alert us to changes in system dynamics. Caspari and Caspari address thisaspect particularly well (10).

What then is an acceptable frequency for buffer hole occurrence? Well that depends. In a relativesense buffer holes should neither be too frequent nor should they be too rare (8). A more concretesuggestion is something less than 10% (11). Clearly the zone 1 buffer hole operates in only alimited number of cases. It is a fine example of exception reporting. After all we want a robustsystem, not one that must be pampered every hour of the day.

If we take our example of a 9 day constraint buffer we would expect most jobs to be completedbefore 6 days and to be waiting at the constraint 3 days prior to their scheduled operating date. Ofthe remaining few we must go and look for them in the prior operations to ensure that they will reachthe constraint before their scheduled operating date there. If we then look at aggregate data for,lets say 25 jobs with roughly a 5% level of buffer hole occurrence, then that data might looksomething like this;



We will find that most jobs arrived during zone 2 or the orange/yellow zone of the buffer, some willeven arrive in zone 3, the green zone. In this case, however, 2 jobs penetrated zone 1, the redzone. It is the penetrations into the red zone that are our buffer holes.

Buffer holes are viewed from the perspective of the system as a whole.

Local Control/Global Feedback Work Order Performance Measures

From the buffer hole data we can construct a more meaningful measure. In the measurementssection we discussed two local performance measures; unit days late and unit days wait. In a for-profit environment these are known as throughput-dollar-days (late) and inventory-dollar-days (wait). I prefer to add late and wait to these measures because to those who are unfamiliar with theterms it is hard to know what they mean or their significance. And these two measures are soimportant to the overall success of drum-buffer-rope that we cant afford to lose people throughobscure terminology.



The immediate impact of a hole in the buffer is to send the buffer manager in search of theoffending job and ensure that it will indeed make the constraint by the appropriate time. In the firstinstance we are really interested in only one thing, that there is an incidence of a problem a bufferhole and whether the problem requires assistance or not. However, we need to quantify this betteronce it is rectified; was it short by a day or 2 days or by how much. And was it a valuable job to thefirm or not. In other words was it a large or valuable job that was very late, or a small job that wasjust a fraction late to the zone 1 buffer. Multiplying the final lateness to zone 1 by the throughputgives us the correct answer throughput-dollar-days late. It gives us a measure of the severity

of the problem.

Now lets consider the other side of the equation inventory. Have you ever had material go to theoutsourcers that seemed to have a life of its own I mean it didnt seem to want to come back. Iknow we said this week, but one of the guys is off and now it will be next week. Even small jobs atthe outsourcer can have major implications if it stops an otherwise completed job from beingshipped. Of course we will pick that up in our buffer holes and in terms of the measure ofthroughput-dollar-days late. However, sometimes, the response is to add some additional just-in-case work to guard against this in the future. Sometimes people will also admit new work in orderto keep a particular center busy. Sometimes the post-constraint areas arent always diligent aboutkeeping work moving. We can guard against these types of incidences with our other localmeasure inventory-dollar-days wait. Adding or building work that may still be on-time iseffectively captured with this measure. Moreover it can be applied to sub-sections of the wholeprocess. It helps to stop the squirrels in the business from storing inventory just-in-case.

Throughput-dollar-days late and inventory-dollar-days wait are the two measures that effectivelyallow us to monitor the stability of the system, they are local measures with implications for globalfeedback. Just two measures. You couldnt wish for better could you? OK, one measure, somepeople cant be pleased, but we will have to suffer two.

Global Feedback Work Order Buffer Resizing

If buffer holes begin to occur in the red zone significantly more frequently, or indeed less frequently,than our target level, we must do one of two things; (1) adjust the buffer or (2) adjust the sprintcapacity. Of the two, adjusting the buffer is the most rapid and easy to do. We simply change apolicy on buffer size and accept a small increase/decrease in total work-in-process and

manufacturing lead time. We can also use the equivalent unit-day late measure to monitor thisneed.

If a buffer hole trend begins to emerge but is not yet too troublesome, then it may indicate theerosion of sprint capacity in the system and the area of that erosion, or it may indicate some special



erosion of sprint capacity in the system and the area of that erosion, or it may indicate some specialcause a machine or quality or operating problem. The cause should be tracked down, monitored,and rectified if possible. There are definitely cases where buffer holes arise from some innocuouschange in the process, we used a different glue because it was cheaper, which has a downstreamimpact that was not anticipated, yes but we are spending more time on removing glue marks. Emergent buffer holes also indicate potential future constraint points. We will investigate thesefactors in more detail in the page on implementation details.

Buffer management is the monitoring aspect of our implementation and monitoring model. Itprovides both local control and maybe more importantly a global level feedback into our drum-buffer-rope basic configuration. We only tune the basic parameters when the system behavior tellsus that we need to do so. We only do what we should do, and we dont do what we shouldnt do.

If we continue to apply our plan of attack, our 5 focusing steps, then it is unavoidable that at somestage we will elevate our system constraint so that it moves to the market. What are we going to dothen? Lets see.

A Market Constraint

If the bottleneck to the system is the market, then we must treat the market as the constraint andsubordinate the process constraint (now best called a control point) to the market. This means onlymaking things that the market needs and not making the things that the market doesnt need(overproduction). Essentially the drum is now in the market place, but we can visualize this byplacing the drum at dispatch the customers desired due date.



In fact, that is kind of old fashioned thinking. Bear with me for a moment, I will try to explain thismore fully.

You see, the control point, our internal weakest link, isnt really so critical anymore, in fact in the nextsection we will see how to operate without explicit scheduling of the control point at all. While theinternal weakest link is no longer critical, it is still however central to our operation. It becomes ourleverage point.

It is certainly seductive to consider that our leverage point should be where we have placed thedrum; that is at the customers required due date. This is, after all, the point where the internalsystem and the external market interact. This certainly appears to be the place where we canmaximize leverage. We can easily test this in a negative sense by running a few orders overdue. We know what the reaction of the market will be. It might be better to consider this interfacebetween the internal system and the external market as a transfer point. We will look at transferpoints again in the introduction to the supply chain pages.

If we step back in our logic we will find that the characteristics of our due date performance is stilldetermined by our internal weakest link. The greater the difference between the additional capacityof the internal weakest link and the market demand; that is the amount that internal capacityexceeds external demand, the smaller the buffer and the shorter the manufacturing lead time. Conversely, the lesser the difference the greater the buffer and the longer the manufacturing leadtime. It is the characteristic of the timeliness of the system that is important; this is an aspect that wewill explore in further detail later on in the paradigms page. For now lets be certain that it is theinteraction of the internal weakest link within the overall system that determines this characteristic.This is where we leverage the whole system from.

Once the constraint is in the market then we should do our utmost to continuously increase themarket demand while increasing internal capability at the same time so as to keep the constraintwithin the market and to be able to provide that market with a very high level of service. In fact our

service level must be much better than anyone else and it will be!

In a market constrained environment we can continue to use drum-buffer-rope with a market drumand an internally scheduled control point, referred to as traditional drum-buffer-rope, or we can beginto use a more recent development called simplified drum-buffer-rope or S-DBR (12). Simplified



to use a more recent development called simplified drum-buffer-rope or S-DBR (12). Simplifieddrum-buffer-rope is at the heart of the Theory of Constraints make-to-stock solution and that iswhere we are heading next so lets investigate simplified drum-buffer-rope first in our morefamiliar make-to-order environment.

Simplified Drum Buffer Rope

Simplified drum-buffer-rope is an excellent reminder to heed the 5th step of the 5 focusing steps;dont allow inertia to become a system constraint. When most drum-buffer-rope

implementations move the constraint into the market, they continue to protect the internal process, atvery the least, at 2 places, the internal weakest link or control point, and the shipping date. Do westill need to do this? The answer appears to be no when the internal weakest link is working at 80%or less of its capacity to supply the market demand. In this situation it is quite safe to roll the safetyup into one global safety buffer instead of two or more.

How do we schedule such a system?

Well, instead of having a constraint schedule and a shipping schedule we now have only a shippingschedule with a gating process off-set by a full shipping rope length. The schedule is still loadedagainst the capacity of the internal constraint available hours per day, or available hours per week,over the average manufacturing lead time, but the only detailed schedule is for shipping. Morecorrectly the schedule is loaded against up to 80% of the aggregate capacity of the internal controlpoint. A queue of some duration will still naturally build and maintain itself in front of the weakestlink, but it is no longer scheduled.

Lets try and draw this.



The drum moves to a proxy position in shipping (it is really in the market but that is a little hard todraw). We now have only one rope and one buffer. Thus simplified drum-buffer-rope allows forfewer buffers without less control and without less protection for the overall system. It also avoidsthe need to determine a detailed schedule for the internal constraint leave it to the operators of theconstraint/control point to determine their own local schedule in accordance with the shippingschedule. Buffer management is still in operation, but now there is only one buffer to manage theshipping buffer.

Avoiding the need to determine a detailed schedule for the internal constraint/control point isespecially useful in situations where capacity is made of up multiple similar machines somepermanently set-up for certain ranges of jobs, some for large batches, some for small batches, andso forth. Another case might be special dependencies or set-ups. In this situation the foremen andtheir people will know better than any planner how best to exploit the system. Leave them to it, theyknow what to do.

There is one further alteration to the basic drum-buffer-rope concepts that we must address and thatis the buffer itself. Simplified drum-buffer-rope was initially described with a two-zone buffer;comprising a green zone and a red zone. However the authors, being pragmatists, have sincerevised this to the basic 3 zone buffer as we have used throughout here. So, lets redraw the systemwith a 3 zone buffer.



The length of the red zone in simplified drum-buffer-rope is defined as the time required forexpediting a medium-sized order. The total buffer length can therefore be sized accordingly. Bufferresizing rules are the same as for drum-buffer-rope.

Simplified drum-buffer-rope, although developed as a consequence of make-to-order systemsmoving towards the desired position of being market constrained, is an ideal system for operating amake-to-stock operation. In the following sections we will look at, firstly, a transitional system thatuses full traditional drum-buffer-rope for make-to-stock and then, secondly, develop the argumentfor using simplified drum-buffer-rope first in make-to-stock and then in make-to-replenish. Rightnow, however, we must introduce the concept of stock buffers.

Stock Buffers

There are two places where the supply chain interacts with manufacturing at the beginning of theprocess and at completion. It isnt much good to have excellent on-time service for make-to-order ifwe dont have the necessary raw materials on hand to begin with, nor is it very useful in make-to-stock if we dont have sufficient resultant stock-on-hand at all times. Often we dont even think of rawmaterials (or inwards goods) and finished goods as a part of the supply chain, they seem, and infact they are, integral to the manufacturing process. But by their very nature they are the end of onesupply chain, and the beginning of another.

In supply chain each and every node for each and every stock becomes its own buffer. Each nodemust contain sufficient stock to meet demand (and variation in demand) during the period from thebeginning of one re-order/resupply cycle to the next. In both make-to-order and make-to-stockenvironments a stock buffer for raw material input occurs at the beginning of the process. Inmanufacturing make-to-stock environments a stock buffer for finished goods output also occurs atcompletion. Lets add these stock buffers to our diagram.



We recognize that in drum-buffer-rope make-to-order both the constraint buffer and/or shippingbuffer are unique in that they are measured in units of time. In contrast the raw material stock bufferand the finished goods stock buffer are measured in units of quantity. This is an interesting andimportant distinction, although they are defined on the basis of time, they are measured in units ofquantity. We will look at this in more detail in a moment, and then we will continue to examine theconcepts that we must apply to finished goods in make-to-stock and make-to-replenish. Later wewill return to look at how we can modify these same concepts to accommodate the more exactingcase of inwards goods in all of these environments.

Why Then Is The Stock Buffer Size & Activity Determined By Quantity?

To understand why we buffer these stock positions just prior to, and just after, manufacturing in unitsof material rather than units of time we need to consider how we determine both the stock bufferactivity and the stock buffer size.

Lets examine stock buffer activity first.

Why do we use units of material for stock buffer activity when we use units of time for the constraint,control points, and shipping buffer activity? The reason is that now the number of units is invariable

but the amount of time or demand that they cover is variable. Sometimes demand is high and weutilize the same number of units faster and over a shorter period. Sometimes demand is low andwe utilize the same number of units more slowly and over a longer period. But a unit is still a unit. Thus;

At a stock node a unit is a unit but the hours may differ



At a stock node a unit is a unit but the hours may differ

The selection of units of time or units of material isnt arbitrary; it is based on the nature of the stepwe wish to protect. If time is invariant (an hour is an hour) then we use time. If quantity is invariant (aunit is a unit) then we use quantity.

Maybe it is much simpler to say that;

We protect stock with a stock buffer

What about the stock buffer size then?

The unique perspective brought about by the designation of a stock node allows us to define thelength of the stock buffer in time. Essentially the buffer is sized and sees a duration that extendsthrough one period of the re-order/resupply cycle. However, the buffer now sees uncommitteddemand we can not tell how much we will sell in the next hour or the next day or whenever. Therefore, we must once again substitute non-variable units of stock in place of the variableamount of time or demand that they cover. Thus the replenishment buffer size is also measured inunits of quantity. We will return to the part the re-order/resupply time plays in determining the stockbuffer size soon.

Transition to Make-To-Stock

In many make-to-stock systems it might be inelegant but never-the-less quite effective to give make-to-stock orders a due date for completion as we would in make-to-order jobs and to thereforeoperate a shipping buffer. The shipping buffer then supplies a stock buffer. The stock buffer itselfwill also contain some measure of safety. This undoubtedly overdoes the safety aspect, as we havesystem safety in safety stocks and we have system safety in the shipping buffer. However, so longas this is recognized, and so long as it works, then it shouldnt be a problem. Lets have a look atsuch a re-order system using full drum-buffer-rope.



Most often the stock levels in the stock buffer of this system will use some variation of aminimum/maximum or re-order point system. Stock tends to be made as needed in set quantities(batches) or multiples of these set quantities. This is a fixed-quantity variable-frequency re-ordering system. Very crudely we could think of this as filling irregularly but completely from the

bottom towards the top of our stock buffer.

In traditional reorder approaches to make-to-stock using min/max systems we can wing a reordersystem using recent rates of consumption to give us a time remaining until order release. Orderswith the least time remaining get priority.

Priority = [(Stock-On-Hand + Work-In-Process Orders) Reorder Point] / Consumption

Such an arrangement of course also allows for a very effective mixed make-to-order/make-to-stockoperation. Probably the greatest majority of businesses currently operate in this manner.

This transitional example is shown with an internal constraint. In this case we can try and supply allof the demand but it is not possible. Therefore, either we must accept that some of the stock bufferswill be out-of-stock some of the time, or we must reduce the total number of stock items just as wemust turn down some orders in an internally constrained make-to-order system. Despite theprotests of purists, make-to-stock systems often start from this position but it isnt so easy torecognize.

In an internally constrained make-to-order system the turning down of orders is often active; wemake a decision to do so. In an internally constrained make-to-stock system the decision issometimes passive; the customer makes it for us. Moreover the customer may never voicedisappointment at a stock-out, they may simply go elsewhere and we might not ever know.

Really, we want our initial internally constrained make-to-stock operation to move towards being



Really, we want our initial internally constrained make-to-stock operation to move towards beingmarket constrained. Think about it. We need sufficient stock so that our customers can always findwhat the want, and when they want. Such market constrained systems tend to be replenishmentsystems which we will discuss in a moment. The objective of this transitional and traditional drum-buffer-rope stage is to make available the needed capacity to get to the required level of output. The identification, exploitation, and subordination steps of full drum-buffer-rope will uncover existingcapacity that was previously unrecognized.

We need to understand why this transitional stage is so common. One reason without doubt is thatmaterials requirements planning (mrp) and material resource planning (MRP II) use time as thedefault for treating stock orders. This is so common that we dont even challenge the assumptionsbehind it. To do so however, we need to introduce a new concept.

We need to introduce the concept of replenishment stock buffers.

Drum-Buffer-Rope & Make-To-Replenish

As we move from being internally process constrained, to being externally market constrained anormal consequence of implementing drum-buffer-rope then there is an opportunity to move fromfixed-quantity variable-frequency make-to-stock to something superior. We will call this superiormode of operation make-to-replenish as distinct from make-to-stock.

First, however, we need to be aware of a subtlety. We need to remind ourselves of our plan ofattack, our 5 focusing steps. They are;

(1) Identify the systems constraints.

(2) Decide how to Exploit the systems constraints.


(4) Elevate the systems constraints.

(5) If in the previous steps a constraint has been broken Go back to step 1, but do not allow

inertia to cause a system constraint. In other words; Dont Stop.

We know that the constraint is in the market, so where then is our leverage point? Using theterminology of Dettmer (13) it has to be somewhere within our span of control rather than our sphereof influence. Just as the due date was a seductive place to consider leveraging from in make-to-order, so too here is the finished goods stock. After all, we must have the right material in the right

place at the right time always.

However, just as with due dates in make-to-order, the greater the difference between the additional

capacity of the internal weakest link and the market demand; that is the amount that internal capacityexceeds external demand, the smaller the queue in front of the weakest link and the shorter themanufacturing lead time and thus the smaller the stock buffer that is required. Conversely, thelesser the difference the greater the queue in front of the weakest link, the longer the manufacturinglead time, and thus the greater the stock buffer that is required. Again, it is the characteristic of thetimeliness of the system that is important and it is the interaction of the internal weakest link with the



timeliness of the system that is important and it is the interaction of the internal weakest link with theoverall system that determines this characteristic. It is from the internal weakest link that weleverage the whole system.

Does that seem logical? Then lets move on.

In a make-to-replenish environment, stock provides the system safety against both the variation indemand from the customer and variation in the time to needed to re-order and resupply the stock. Therefore we dont need to use a shipping buffer; the stock is our buffer.

This is an ideal system, lets see how it looks, lets draw this new system.

This is a replenishment system. It is no-longer internally constrained, the drum has therefore

moved to the market. The finished goods replenishment buffer is the sole buffer in the system foreach stock unit that we produce. At a pre-determined frequency; hourly, daily, or weekly, we mustcheck and raise a material release order sufficient to replenish the buffer back to its full level. Thisis a fixed-frequency variable-quantity replenishment system. Very crudely we could think of

this as filling regularly but somewhat incompletely from the top towards the bottom of ourreplenishment buffer.

Replenishment buffers characteristically oscillate between being nearly full and being somewhatempty. They oscillate between zone 3 and zone 2 rarely do they move into zone 1. Movement intozone 1 is a signal for management action (action by exception). If having 1/3rd of your stock sittingin zone 1 sounds wasteful remember that this system is frequency driven. Total finished goodsstock may well be 1/2 or 1/4 or even less of that previously required at this first stage aftermanufacturing. Lets check that we understand why.



In the earlier discussion on drum-buffer-rope make-to-order we limited ourselves to the benefits thatoccur when excess work-in-process is removed from the system, we dried the system out by atleast 50%; and lead time was reduced proportionally. We didnt consider the what if when anorder consists of 1000 units of the same thing, surely we could manufacture that as two lots of 500? In a way we would have been getting ahead of ourselves, in fact the whole subject of batching is soimportant that there is a further page on it called batch issues. However, because batching isendemic to make-to-stock we need to address it here to some extent.

So lets ask the what if. What if we have a batch of stock of 1000 units, and what if we made twosmaller batches of 500 units instead? Well, each batch would move through the process twice asfast as before. The rationale is that most of the time that a batch spends on the floor is time spentwaiting, and therefore a batch of 500 waits half as long as a batch of 1000. Unconvinced about thewaiting? Go attach a balloon to a batch or two and see what they are doing over a couple of days.

So, if we dry the system out of excess work-in-process and that reduces lead time by at least half,and then we reduce batch size in-turn by half that will reduce lead time to a quarter without eventrying. Of course we have to run those batches twice as often, could that be a problem? Well notusually. It is not a problem as long as we dont turn a non-constraint into a constraint by too manyset-ups.

Now lets look at two further considerations.

Firstly; we dont batch when we replenish! Oops. Trying not to use the term batch from now on

is going to be so darn difficult, however, it is not politically correct to talk about fixed-batches inreplenishment because they dont exist, therefore we will talk about the variable replenishment

aggregate size instead. As Schragenheim points out, in all of the standard approaches to make-

to-stock there is a buried assumption somewhere as to a fixed batch size (8). We assumed exactlythat in our transitional approach above. So yes, common sense tells us that we must aggregatesome of the stock units some of the time, least we grind the system to a halt, or we are Toyota, orwe are very clever. However, we should strive to keep the size of our replenishment aggregate assmall as possible and variable; that is, not invariably the same size each and every time.

Secondly; we may have product cycles. Sometimes when the nature of the material is differentfor different stock we may wish to make everything required with one material before we change toanother. Or we might have a preferred sequence of set-ups either up or down the size range orsome other set of dependencies that cause us to cycle through our manufacturing process. Thegood news is that cycle times will decrease in proportion to our variable replenishment sizereduction.

Why is this so important?

It is important because our initial stock buffer size, and hence the physical inventory and financial

investment that we must carry, depends upon the resupply time.

In the absence of product cycles and large variable replenishment aggregates the manufacturinglead time determines the resupply time.

In the presence of product cycles, the time for the completion of a cycle plus the manufacturing leadtime determines the resupply time.



time determines the resupply time.

There is a more detailed explanation of the characteristics of replenishment in the second part ofthe replenishment page in the supply chain section. Be aware that manufacturing professionalstend to misunderstand replenishment at first unless they have been exposed to supply chain

management. If we have only ever been exposed to fixed batches of material then everything tendsto look like a fixed batch of material. This is a very real and substantial block. Please check thereplenishment page to test your understanding.

During the implementation phase of simplified drum-buffer-rope in a make-to-replenish environmentthe removal of fixed batch sizes and replacement with smaller more frequent variable replenishment

size orders is the critical parameter in determining system behavior.

Local Planning In Make-To-Replenish

If the demand in make-to-replenish is uncommitted, then how do we plan? Well, the same way as inmake-to-order simplified drum-buffer-rope, we load replenishment orders against our aggregatecapacity our daily or weekly capacity multiplied by the material release rope length. The capacityis defined as that of the weakest link in the process. We shouldnt fill more than about 80% of thatcapacity over any medium term plan. Lets add this to our model.



We will see prioritization rules for loading orders in the discussion on local control. Although stockorder prioritization is a local planning action rather than local control, we need to first understandbuffer status in order to understand make-to-replenish more fully than we do at this point.

An Initial Finished Goods Stock Buffer Sizing Rule

The amount of stock that we must hold in our stock buffer depends upon the amount of time that itmust cover for. The principle components of that time are the reorder time and the resupply time. We can define the stock buffer as follows;

Finished Goods Stock Buffer = Average Demand x (Average Re-order & Re-supply Time)+ A Margin of Safety

Because we use averages for demand and for time we use the margin of safety to accommodatethat which we know least well, the variation around the average values. If the resulting buffer is toobig it will soon become apparent and we can, in fact we must, resize it. Likewise, if the bufferproves to be too small then we will also have to resize it.

Re-order and resupply time are clearly the most critical components. We have already discussedthe factors that lead to a reduction in resupply time, most of which are within our own control; whatthen are the issues that determine the re-order frequency?

Policy of course!

Whose policy?

Our policy!

Yep, there are not a whole lot of excuses why re-order, which is totally under our own control in ourown system, can not be substantially reduced in duration. If we are big enough to have a realproblem then we will have a computer and that can easily be interrogated more frequently. If we are



problem then we will have a computer and that can easily be interrogated more frequently. If we aresmall enough to have no problem then we just need to do the sums more frequently.

Buffer Management Make-To-Replenish

In drum-buffer-rope make-to-replenish, buffer management once again provides us with both localcontrol via buffer status and global feedback in the form of buffer resizing via longer term trends inbuffer behavior. The model is the same; the terminology is modified to suit the environment.

Lets then look at how we determine buffer status, stock order status on the floor, and new stockorder release priority. Then we are in a position to evaluate the longer term performance measuresused for buffer resizing.

Local Control Stock Buffer Status

Stock buffer status is defined in the same way as a time buffer;

Stock Buffer Status = (Buffer Quantity Stock-On-Hand) / Buffer Quantity

We have already used this concept with buffers for work orders in make-to-order, exactly the sameprinciples apply for stock buffers in make-to-replenish. Lets illustrate the concept with a fewdiagrams for stock buffers. The following buffers contain varying amounts of stock-on-hand (SOH).



Buffer A has been depleted by 58%, buffer B by 75%, buffer C by just 17% and buffer D by 50%. Clearly the buffer status of buffer B indicates that we should investigate and if necessary expediteany in-process stock orders.

Of the other buffers; C is in the green zone, zone 1, and of no concern, buffers A and D are within theyellow/orange zone, zone 2, and we should leave well alone unless we once again wish to tamperwith the system and Deming taught us what happens when we allow that occur. So lets not do it.

By using buffer status we can prioritize different product buffers amongst each other regardless ofwhether one replenishment buffer covers a period of 2 months and 2000 units and anotherreplenishment buffer covers just 2 days and 20 units.

Local Control Stock Order Status

In make-to-order the relative priority of released work orders can only be affected by internalthings, however, in make-to-replenish the relative priority of stock orders can also be affected byexternal changes in the market demand during the processing time. Because stock orders shouldbe smaller and more frequent it is also possible that more than one order for identical stock can bein-process at the same time, this too must be taken into consideration.

Thus the stock order status for material on the floor is (8);

Stock Order Status = (Buffer Quantity Stock-On-Hand WIP Ahead) / Buffer Quantity

Lets illustrate this with two different stock items light blue and lavender.



The buffer status of lavender is 84%, it is well into the red zone, and we should therefore ensure thatthe lavender stock order #1 has priority over other stock orders such as the light blue stock order#1. What about lavender stock order #2 then? This stock order with a status of 50% still has priorityover light blue stock order #2 at 25% but the only place that we might exercise this is at the slowestpoint if both orders are waiting in-queue. The status of released stock on the floor is important andcan change over the term that it is in-process, however what we should do about this depends; itdepends on the status of the finished goods buffer itself. As long as the finished goods buffer statusis not in the red zone we should leave the process alone.

Local Control New Stock Order Release Priority

Buffer status tells us about the end condition of the system and whether to check or maybe facilitatework already released for production. However, without customer due dates as in make-to-orderjobs, how can we prioritize new jobs for material release? To do this we must replenish the buffer toits full capacity allowing for current stock-on-hand and current work-in-process, thus;

Stock Release Priority = (Buffer Quantity Stock-On-Hand Total WIP) / Buffer Quantity

For the example that we just used with lavender and light blue finished goods stock buffers, wehave the following release priorities.



Both the light blue and the lavender stock have a release priority of 9%. In a replenishmentenvironment release priorities are not likely to become very high, it is the relative value not theabsolute value that is of most importance.

The actual amount of material released to replenish the buffer will be the amount required to bring itbe back to the full buffer quantity;

Stock Replenishment Quantity = Buffer Quantity Stock-On-Hand Work-In-Process

When demand is generally high, then the time between one stock order being released and the nextorder for the same stock will increase because each order is now larger and takes longer toprocess, however overall system set-up will decrease as a consequence. Conversely, whendemand is light, the time between each order will decrease because each order is now smaller andtakes shorter time to process. Overall system set-up will increase as a consequence. Schragenheim argues persuasively how this mechanism self-adjusts to load and can be used todrive system stability (8).

Do you notice something important? We have moved away from fixed batch sizes and away fromthe fixed schedules of transitional make-to-stock. The re-order frequency and resupply frequencyand quantity are now totally determined by the characteristics of the system that we implement andthe demand characteristics of the customers. This is make-to-replenish.

In fact we are not entirely free of fixed schedules. In replenishment we have at the very least;

rechecking, re-ordering and resupply. The re-ordering and resupply are indeed free from fixed-frequency scheduling but there is one last hold-out the rechecking. Most likely we will still batchthis in time to once a day, towards the end of the day maybe when we know what we have soldthat day.

An Alternative Stock Buffer Sizing Rule



There is an alternative stock buffer sizing rule for finished goods stock buffers. Schragenheim andDettmer use the term emergency level to define zone 1; they suggest that zone 1, the red zone oremergency level, should contain sufficient stock to cover the time required to expedite a mediumsize replenishment order through the system (14). Three times this value will therefore define the fullbuffer size.

Lets make sure that we understand all of the terms that