Costs for running scientific applications on EC2 significantly higher in comparison to in-house infrastructure

We have recently written a scientific article on “Towards a Model for Cloud Computing Cost Estimation with Reserved Instances”.  Cloud computing has been touted as a lower-cost alternative to in-house IT infrastructure recently. However, case studies and anecdotal evidence suggest that it is not always cheaper to use cloud computing resources in lieu of in-house infrastructure. Also, several factors influence the cost of sourcing computing resources from the cloud. For example, cloud computing providers o er virtual machine instances of diff erent types. Each type of virtual machine strikes a diff erent tradeoff between memory capacity, processing power and cost. Also, some providers off er discounts over the hourly price of renting a virtual machine instance if the instance is reserved in advance and an upfront reservation fee is paid. The choice of virtual machine types and the reservation schedule have a direct impact on the running costs. Therefore, IT decision-makers need tools that allow them to determine the potential cost of sourcing their computing resources from the cloud and the optimal sourcing strategy. The paper presents an initial model for estimating the optimal cost of replacing in-house servers with cloud computing resources. The model takes as input the load curve, RAM, storage and network usage observed in the in-house servers over a representative season. Based on this input, the model produces an estimate of the amount of virtual machine instances required across the planning time, in order to replace the in-house infrastructure. As an initial validation of the model, we have applied it to assess the cost of replacing an HPC cluster with virtual machine instances sourced from Amazon EC2.

After running the optimization over the scenario mentioned, we have got the following results. Off ering the same kind of service on AWS as the NICPB WNB cluster does, it would cost roughly $980k. These costs comprise of $844k for CPU usage, $13.415 for data upload, $1.579 for outgoing communications and $120k for data storage. These numbers illustrate the fact that the major operational costs come from the usage of EC2. Internal accounting of the in-house infrastructure showed that these cloud operation costs are significantly higher than the equivalent in-house costs. Note also that we have not even included storage migration cost into the cloud. With storage migration the network costs would be much higher as the data exchange within cloud is free. With the average storage space of 90TB moving in the data would incur an additional cost of $5-6 thousand plus the price of the 45 2TB hard disks.

In a similar analysis of running scienti c computing application on clouds, people observed that the cost of running an application on the public cloud depends heavily on the usage pro le of the application. Our model shows that if we use cloud-based storage, the main expense for the HPC applications on Amazon cloud is the computational part.

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.

4 Responses to “Costs for running scientific applications on EC2 significantly higher in comparison to in-house infrastructure”

  1. I was looking forward to reading this paper, but it doesn’t seem to be online yet. I hope it cites the following paper, which made much the same observation a couple of years ago:

    Free Factories: Unified Infrastructure for Data Intensive Web Services, Alexander Wait Zaranek, Tom Clegg, Ward Vandewege, and George M. Church, Harvard University. USENIX’08. http://www.usenix.org/events/usenix08/tech/full_papers/zaranek/zaranek.pdf

    There are, of course, other advantages to cloud computing beyond the simple cost aspects. Availability, disaster resiliency, and flexibility all come to mind.

  2. The key obstacle in assessing the cost of a cloud offer remains the lack of an industry wide measure of cloud processing capacity. We have GB for memory, TB for storage, and GB for bandwidth.

    The Cloud Price Calculator http://cloudpricecalculator.com addresses this by adopting Amazon’s ECU as the compute metric at 1ECU = a 400 Passmark score.

    Combining all the resources and dividing by price yields the Cloud Price Normalization index and a ranking of cloud offers. Interesting, the ranking shows Amazon’s newer instances provide more value than the older ones as Amazon has rarely reduced prices after introducing an instance.

Trackbacks/Pingbacks

  1. Research group digest – ulno.net - March 18, 2011

    […] Costs for running scientific applications on EC2 significantly higher in comparison to in-house infr… We have recently written a scientific article on “Towards a Model for Cloud Computing Cost Estimation with Reserved Instances”.  Cloud computing has been touted as a lower-cost alternative to in-house IT infrastructure recently. However, case studies and anecdotal evidence suggest that it is not always cheaper to use cloud computing resources in lieu of in-house […] […]

  2. Cloud Computing – Amazon, Google & Co against HP, IBM & Co | Cloud Computing Economics - March 24, 2011

    […] not be the case for every company and for every application (see also blog post about the costs of running scientific applications on the cloud and the post about how WAN costs influence cloud ROI) . And of course not every business will be […]

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: