Top Banner

of 9

Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

Aug 08, 2018

Download

Documents

Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    1/9

    DatacenterConsolidationClient Story

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    2/9

    Summary

    The client, an iconic, global nancial services business headquartered in the US, hasannual revenues of $30bn and 60,000 employees.

    The client turned to Sumerian for decision support related to integrating a datacenterinherited from a signicant acquisition. Sumerian was asked to apply its technology

    and services to fully assess evidence-based options for datacenter consolidation.Sumerians recommendations allowed the client to break away from its traditional liand drop methodology for the rst time.

    Sumerians predictive analytics technology and Data Scientists provided statisticalbased evidence that supported the clients decisions by dening, exactly, the right-sized and optimized target environment for consolidation from 3 to 2 datacenters.

    The client achieved a total cost saving of approximately $1M per annum by usingSumerians analytics to accurately predict and optimize the capacity required in theretained datacenters. This was in addition to the $3.2m saving directly attributable to

    closing the acquired facility.To see an example report for the type of analytics conducted to support this client,please refer to our Datacenter Consolidation sample report.

    1 The challenge

    1.1 The client was planning an IT cost reduction program. A prime option was torationalize the number of datacenters they operated. What would the eectbe of closing a recently acquired regional datacenter? It had been identiedthat such a closure would save the client approximately $3.2 million per annumin operating costs. However, an additional opportunity, and challenge, was tofurther increase savings by an eective use of capacity in their 2 well establishedUS datacenters.

    1.2 The clients initial plan was to adopt a very conservative approach transferringthe target regional datacenters assets on a like-for-like basis into the 2 main USdatacenters. However, they suspected that this approach might leave signicantpotential cost savings and eciency gains on the table. The ideal result wouldbe to transition the regional datacenter workload to spare capacity currentlyhidden in the 2 datacenters to be retained. If this could be proved as achievable,it would minimize the amount of expenditure on new hardware and reduceoperating costs.

    1.3 The critical success factors which the client identied as essential to enable themto maximize cost savings were:

    a) Quantify current over-provisioning in the regional datacenter and predict aright-sized target environment (in contrast to like-for-like provisioning).

    b) Predict the amount of capacity reduction that could be achieved, whilst stillmaintaining current service performance and availability levels.

    c) Identify the spare capacity in the existing 2 US datacenters and use this tocreate the target environment rather than buy additional new capacity.

    1.4 The client engaged with Sumerian to do this accurately and to provide theevidence to prove this potential option.

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    3/9

    2 Sumerian solution

    2.1 Sumerian applied the combination of Workbench, its analytics platform, andthe specialized skills and experience of its Data Scientists to quantify thesefactors and to recommend an optimized target environment.

    2.2 Using evidence-based analytics, Sumerian was able to tailor a more cost

    e

    ective strategy. By measuring the existing regional datacenter utilization levels,Sumerian predicted the actual capacity required. Then, using our predictivescenario modeling capability, Sumerian modeled the lowest cost options toaccommodate the applications and data storage from the regional datacenterin to the remaining 2 US datacenters. This approach minimized the need foradditional capital expenditure and maximized future operational cost savings.

    3 Sumerian method

    3.1 Sumerian utilized its proven methodology to generate these results, with therecommendations delivered 4 weeks aer the initial data was uploaded.

    3.2 Step 1: Collect data

    First, Sumerian collected and analyzed historic data for the relevant datacenterinfrastructure platforms, and used this to understand the current usage levelsand consumption proles.

    3.2.1 Aer an initial discovery phase working on-site with the client, Sumerianidentied 3 sets of system data to be used for this analysis; CMDBinventory, SCOM systems utilization logs and SiteScope performance logs.This data already existed within the clients estate and was, therefore,readily available with minimal intrusion and disruption. It was quickly andsecurely established in the Sumerian datacenter for analysis. This data was

    then integrated into existing Sumerian models, and the initial results werepresented in a format that was easy for the client to assimilate.

    3.3 Step 2: Create baseline

    In this step, Sumerian created an overall baseline workload prole for the entiredatacenter server and storage estate and predicted the actual capacity(right-sized) required to support this workload. This was used to create a highlevel overview of current datacenter utilization levels and to show infrastructureconsumption over time; at monthly, weekly, daily and hourly levels. Analysisover an extended period of time was key to identifying any short duration busy

    periods as well as any cyclical variations in consumption patterns. This highgranularity of data all needed to be factored into the overall workload sizing.

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    4/9

    3.3.1 The CPU and memory utilization prole for each server was expressed usingboth percentiles and the maximum values observed during the analysistime period. A percentile is the value of a given metric below which aproportion of observations fall and is an established method of representingserver utilization proles over an extended measurement period.

    3.3.2 If the 95th percentile CPU utilization for a server is 60%, this means that 95%

    of the CPU utilization values recorded are below 60%. Below are someexamples of how the overall estate storage and server utilization levelswere visualized.

    Fig 1- Overall server storage used

    Fig 2 Virtual server CPU utilization expressed as 75th and 95th percentile and maximum, memory

    shown as maximum.

    3.3.3 From the above diagram (Fig 2) one can see that, for the virtual machines,the overall 75th percentile CPU utilization is 5.8%. The overall memoryutilization (based on the maximum) is only 25% out of the total physicalmemory of 100GB. This indicates a signicant opportunity for consolidationand capacity reduction in the target environment.

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    5/9

    Fig 3 Physical server CPU utilization expressed as 75th and 95th percentile and maximum, memory

    shown as maximum

    3.3.4 From the above diagram (Fig 3) one can see that, for the physicalmachines, the overall 75th percentile CPU utilization is 9.3%. The overallmemory utilization (based on the maximum) is 56.5% of the total physicalmemory of 1200GB. As for the virtual servers shown above, this indicates asignicant opportunity for consolidation and capacity reduction.

    3.3.5 In contrast to these aggregated utilization views, this data was alsoused by Sumerians Data Scientists to undertake a further deep diveanalysis of each server and its associated applications, and present aneven more granular view of usage over time. This allowed each serverand application to be accurately proled and complete and accurateconsumption proles to be created. An example of how this data was

    presented is shown below (Fig 4). In this example, one can see the actualprole for each of the top applications measured over a 3 month period.

    Fig 4 - Top application usage across the virtual server estate, again expressed as percentile values

    and maximums. (Each line represents a server, but the identication of each sever, usually shown to

    the le, has been hidden.)

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    6/9

    3.4 Step 3: Model workload and predict capacity requirements

    The next step was to model the current workload in the target environment

    and calculate the actual capacity required using the key system metrics. While

    many hundreds of metrics were available in the system data being analyzed,

    for this task Sumerians Data Scientists focused on the most relevant metrics that

    inuence resource consumption: in the case of servers this is CPU and memory

    utilization. The measured CPU and memory utilization gures were used in

    conjunction with the server inventory information to predict the right-sized target

    environment.

    3.4.1 The initial Sumerian baseline analysis identied that there was likely to be

    signicant potential for consolidation in the target deployment. Whilst

    there were periods of high levels of utilization, these periods were transient

    and, therefore, it would be appropriate to size on the basis of the 75th

    percentile of utilization with the shared resources of the virtual environment

    being utilized as required.

    3.4.2 This analysis built upon the utilization summary calculated previously by

    relating the utilization gures to the server inventory information for server

    builds, to ascertain the sizing requirement for CPU in MHz and for memory

    in GB. These gures could then be compared with the client specic

    reference architecture in the target environments to identify the server

    compute and memory size required.

    3.4.3 It was recognised that servers have dierent daily proles based upon

    utilization patterns over the period observed. These proles were

    statistically modeled using a technique called polynomial regression. This

    ensures that the utilization proles do not exceed the capacity of thetarget machine at any given hour of the day.

    3.4.4 These models were tted using the hourly 75th percentile, 95th percentile

    and maximum to represent diering risk of exceeding the machine

    capacity and consequent usage of shared resources. The polynomial

    regression models measured both the trend in overall utilization pattern

    and also the volatility around this trend. A condence interval was

    estimated around the polynomial curve tted to the data: this varies

    according to the volatility in the server prole.

    (see Fig 5 on the next page)

    3.5 Step 4: Optimize using What if? scenario modelings

    To show how further consolidation could be achieved by application level

    optimization, Sumerian used what if scenario modeling to identify where

    stacking or sharing applications across a reduced number of server instances

    could be applied, factoring in considerations such as: security policies,

    availability targets and business rules. Once optimal models were created, these

    could be used to recalculate and predict the new capacity requirements.

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    7/9

    Fig 5 - Example of the polynomial curve showing volatility.

    Fig 6 An example of how application workloads can be modeled to show how these could be

    stacked on single server instances and remain within the available headroom capacity. Application

    names have been removed from the empty le hand column.

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    8/9

    3.6 Step 5: Create cost models and calculate cost savings

    Finally, Sumerian created cost models to provide indicative costs for alternative

    optimization models and the costs savings that could be achieved through

    workload and application optimization.

    3.6.1 Cost estimates and comparison were presented based on the following

    criteria:

    a) Like-for-like cost where the existing datacenter deployment is mapped

    to equivalent server conguration in the hosted environment. This is

    equivalent to the datacenter server inventory with no resource utilization

    applied.

    b) Optimized cost where the existing datacenter deployment is mapped

    to VMs in the hosted environment, based on resource utilization and the

    stacking of applications on VMs.

    4 Outcomes and results

    4.1 Sumerian delivered results to the client 4 weeks aer the initial data was

    uploaded. By right-sizing the measured workload requirements for the

    acquired regional datacenter, Sumerian was able to accurately predict the

    actual capacity required to support the current application workload, and

    therefore specify the exact sizing and capacity requirements in the 2 retained

    datacenters.

    4.2 Sumerian provided a plan for optimal application deployment to ensure that

    all available headroom was utilized as eectively as possible, giving maximumeciency and lowering ongoing running costs, at the same time as protecting

    application performance and availability goals. The outcomes of Sumerians

    work were:

    a) Quantied annual costs savings of around $0.5M per annum, achieved

    through right-sizing and optimizing the current workload in the Production

    Virtual Server estate.

    b) Additional savings from moving the existing Test and Development

    workloads into spare capacity in an alternative datacenter, estimated at

    $0.3M per annum.c) Further savings in the associated operational support, licensing and HVAC

    costs, achieved by the reduction in the overall size of the server estate and

    equating to an additional $0.2m saving per annum. This was equivalent to

    approximately a 25% reduction.

    4.3 The client had estimated an operating cost saving of $3.2M from closing the

    acquired regional datacenter. This is not directly attributable to Sumerians work.

    However, Sumerians clear evidence identied additional annual savings of

    $1M. This was achieved due to Sumerian proving that the traditional like-for-

    like approach could be avoided, and Sumerians detailed recommendations

    Copyright 2013 Sumerian Europe

  • 8/22/2019 Sumerian_Client Stories_Full_Datacenter Consolidation.pdf

    9/9

    on how the workloads in the withdrawn regional datacenter could be provided

    within the resources already existing, but unidentied, in the 2 remaining

    datacenters.

    In eect, Sumerians technology and expertise increased the savings identiedby the client itself by a further 30%.

    Copyright 2013 Sumerian Europe

    More information

    To nd out more about our client stories,just give us a call on 0131 226 9300, dropan email to [email protected] or visitour website at www.sumerian.com