Cloud Computing Economics Ville Volanen [email protected]
Dec 22, 2015
Packaged Software There are several fixed costs
Licence Hardware Datacenter Maintanance staff
Scaling is problematic Scaling up can be expensive and require of
expertice and time if you need to invest to hardware
Scaling down might not save much as fixed costs stay the same
Difficult to estimate costs High barrier of entry
OpEx
CapEx
Cloud Service provider builds and maintains the capacity
Hardware Datacenter Maintanance staff
Customer pays what s/he uses Scaling is simple
Costs of customer depend on usage No investment to hardware required Service provider has less need for scaling (average
usage) Costs are easy to estimate Low barrier of entry
OpEx
Datacenter, traditional way Build datacenter componen by component Each component needs to be handled
separately Ordering and shipping Unpacking, racking, installing Maintanance is continuous
Lots of skilled labour is needed Increasing capacity increases costs
More ordering, shipping, unpacking, maintanance etc
New workforce needed
Datacenter, better option Build datacenter module by module
Each module with 1000 (or so) systems Modules need only electricity, network
connection and cooling water No unpacking or installing Shipped once (instead of thousand times) and
dropped to place Modules are not serviced during their life span
Due to high number of systems in module broken hardware has only a small impact of module performance
Aged modules are shipped back to manufacturer and recycled
Datacenter, better option, cont Adding capacity has only minimal effect to
costs (as long as there is room) No service, installations, etc. Economies of scale drop costs
Module assembly can be automated Again, economies of scale drop costs
Some extra challenges if the modules need to store information Broken disks are not replaced since there is no
service For example networking does not have this problem
Is cloud always a good option? Would it be economical to buy a couple of
virtual servers for every day use from cloud? Initial costs would be lower but is that relevant?
But also virtual servers need maintanance (install and update apllications, performance monitoring etc). Cost per server could be around €1000 Cost per technician could be around €50.000 each
year
Acquiring a relatively small number of servers from cloud is not necessarily very economical
Applications ”blocks” For example services provided by Amazon and
Google for system development Simple databases
http://labs.google.com/papers/bigtable.html Data storage http://aws.amazon.com/s3/ …
Developing similar capabilities from scrath would cost huge amounts of money (datacenter costs plus software development)
Allows faster time to market as some blocks are already there. Also development can be done in production-like enviroment.
Internal services as cloud Typically cloud computing is seen as a service
for external customers Company internal clouds are possible as well.
IT department is seen more like an internal service provider
This increases control to resources and adds accountability
Direct cost savings smaller compared to external clouds due to smaller number of ”customers” Economies of scale
Conclusions Cloud computing can potentially create huge
savings Calculations from analysts suggest savings up to 80%
in business applications are possible Cost profile is different between cloud and
traditional computing CapEx to OpEx
Cloud computing is not always cheaper In addition to dropped unit costs cloud can reduce
barrier of entry and time to market
References Architecture for Modular Data Centers
http://arxiv.org/ftp/cs/papers/0612/0612110.pdf
The Benefits of Virtualization and Cloud Computing (http://virtualization.sys-con.com/node/870217 )
Cloud computing economics (http://www.cloudcomputingeconomics.com/)
http://en.wikipedia.org/wiki/Cloud_computing