To cloud or not to cloud as a function of data center utilization

The economics of cloud computing is a topic that is far less trivial than it looks at first sight. Despite cloud computing being one of the big buzzwords in IT these days, literature on cloud computing economics is very rare.

We have screened a significant part of the research on cloud computing and selected  three papers to take a closer look at. The paper that gives the best overview of the topic is the one by M. Armbrust et al., “Above the clouds: A Berkeley view of cloud computing” from the UC Berkeley Reliable Adaptive Distributed Systems Laboratory. The authors mainly focus on cost saving and risk aspects in their economic reasoning. Businesses are constantly faced with the challenge to estimate the expected peak resource utilization of their data centers and dimension them accordingly. This poses the risk that the decision makers might either overestimate the demand with the consequence that the data centers will be under utilized. If they on the other hand underestimate the demand, enterprises might end up losing potential customers by not being able to serve them during peak hours.  Case studies show that the average utilization in data centers is around 5% to 20%, reflecting the fact the peak workload can exceed the average by a factor of 2 to 10.  Apart from potential direct cost savings by migrating to the cloud, intangible benefits are due to this transfer of risks from the company data center to the cloud computing service provider.  In theory this will be a win-win for both parties. The cloud computing service provider needs a critical amount of customers to utilize his assets whereas his clients can match their computing power purchases with their requirements. The efficiency gain would be shared by both parties as cost savings for the client and additional revenue for the provider.

Ambrust et al. managed to pour this into the following inequality that needs to be fulfilled in order to justify that a service is hosted in the cloud.

UserHourscloud x ( revenue(perhour) – Costcloud(perhour)>=UserHoursdatacenter x ( revenue(perhour) – Costdatacenter/Utilization)

The left side the margin when using the cloud service whereas the right side calculated the margin when using the fixed-capacity data center.  This inequality is of course true but still a decision maker is faced with the problem of foreseeing user hours and data center utilization.

The authors of the paper mention two more arguments for cloud computing. Companies having to decommission hardware (due to an upgrade) that was not fully depreciated are facing a financial loss.  Cloud computing can eliminate the risk for this financial loss . Secondly the authors argue that cloud computing providers can pass on savings on hardware costs more quickly to due to their high purchasing power.

A very important aspect regarding cloud computing is Jim Gray’s (4) conclusion in 2003 that economics make it necessary putting the data center hear the application as cost of wide-area networking keep falling more slowly (and are relatively higher) than all other IT hardware costs.

This paper contains some valuable information regarding the economics of cloud computing.  It gives a sound overview of the topic but it does not dig really deeply. It shows some very simplified case studies but and provides a very simple commercial condition on when the margins of the a cloud service would surpass the margins from a local fixed-capacity data center. They give the following arguments for cloud computing.

  • transfer of  risks (very likeley)
  • cost savings (depends on how well demand is estimated)

Looking at this overview paper from the commercial side,  a lot of questions remain unanswered. The authors manage to point out some interesting aspects clearly. Yet to really make a sound decision on whether to cloud or not to cloud we think that more is needed. We think that there is a need for a holistic, integrated approach similar to a due diligence or a  balanced scorecard.  This approach needs to address the following questions:

How do we weight business strategic aspects against potential cost savings? For some businesses it might make perfect sense to cloud even though it does not financially pay off immediately but in the long run if this migration is in line with their strategy. Other enterprises could maybe save costs quickly but would lose their competitive advantage in the long run.

If we had an integrated economic model for cloud computing, how robust would it be against changes in the cloud computing provider’s price structure? We think that a thorough understanding of the relations needs to be established here in order to be on the safe side with and ROI projections.

How do we integrate the costs of migrating (e.g. customizing software) into a potential model? The costs for an established enterprise to move its IT operations to the cloud might be significant and need to be carefully and thoroughly estimated prior to making a decision.

How do external parameters like company size or industry go into an integrated model? It needs to be investigated whether and to what extent those external not IT related parameters impact a cloud computing model.

How do we evaluate the risks of  a potential migration and integrate them into a model? Looking at what risk a migration to SAP can impose on an enterprise a thorough risk analysis  will be essential. The outcome of this risk analysis will have to be integrated into the model and put into comparison to the potential gains.

How does the cloud architecture (EC2, Azure, AppEngine) influence the decision to cloud or not to cloud? As the paper has nicely shown, the offers vary significantly and are not directly comparable.  So it is not correct talk of “moving to the cloud”. A decision support model will have to reflect the architecture.

We are confident that we can answer some of these questions better after building the first model prototype during the SITIO project.

Sources:

  1. M. Armbrust et al., “Above the clouds: A berkeley view of cloud computing,” EECS Department, University of California, Berkeley, Tech. Rep. UCB/EECS-2009-28 (2009).
  2. Jim Gray, “Distributed Computing Economics,” Queue 6, no. 3 (2008): 63-68, http://portal.acm.org/ft_gateway.cfm?id=1394131&type=html&coll=GUIDE&dl=GUIDE&CFID=71686573&CFTOKEN=70580310.
Advertisements

About Georg Singer

Georg Singer is an analyst at the University of Tartu in Estonia. He works for the institute for computer science on cloud computing economics.

No comments yet... Be the first to leave a reply!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: