The Cost of Running A Service on the Cloud Compared With Running It Locally

My previous posts have described the costs of running data applications on the cloud.  In this post, I provide a simple example of a cost-effectiveness study to answer the question: Is it cheaper to host an on-demand image mosaic service locally or on the Amazon EC2 cloud? The costs described here are current as of October 2010. See Berriman et al. (2010) for a full description.

The cost calculations presented assume that the two services process requests for 36,000 mosaics of 2MASS images (total size 10TB) of size 4 sq deg over a period of three years. This workload is typical of the requests made to an existing image mosaic service hosted at the Infrared Processing and Analysis Center (http://hachi.ipac.caltech.edu:8080/montage/).  We assume that the processing is done on 2.66 GHz dual core Xeon processors with 2 GB memory (designated c1.medium by Amazon) or equivalent machines. These are the most cost-effective Amazon machines for running Montage (see What Types of Science Applications Are Best Run On The Cloud?).

Table I summarizes the costs of the local service, using hardware choices typical of those used at my home institution, the Infrared Processing and Analysis Center (IPAC) at Caltech. The roll-up of the power, cooling and administration are estimates provided by IPAC system management. Table II gives similar calculations for Amazon EC2; the costs there include the costs of data transfer, I/O etc. Clearly, the local service is the least expensive choice. The high costs of data storage in Amazon EC2, and the high cost of data transfer and I/O in the case of an I/O-bound application like Montage, make Amazon EC2 much less attractive than a local service.

The results indicate that for an I/O bound application, which are common in astronomy, the high data storage costs and the high I/O and transfer-out costs may make Amazon EC2 less attractive than a locally hosted application in this example.  Now, this does not mean that Amazon EC2 is a poor choice in general. Amazon would, for instance, become more cost effective as the load on the service declines. In a future post, I will describe how a processing bound astronomy application can be much more cost effective to run on Amazon than on a local machine.

TABLE I.  COST PER MOSAIC OF A LOCALLY HOSTED IMAGE MOSAIC SERVICE

Item Cost ($)
12 TB RAID 5 disk farm and enclosure ( 3 yr support) 12,000
Dell 2650 Xeon quad–core processor, 1 TB staging area 5,000
Power, cooling and administration 6,000
Total 3-year Cost 23,000
Cost per mosaic 0.64

TABLE II.                  COST  PER MOSAIC OF A MOSAIC SERVICE HOSTED IN THE AMAZON EC2 CLOUD

Item Cost ($)
Network Transfer In 1,000
Data Storage on Elastic Block Storage 36,000
Processor Cost (c1.medium) 4,500
I/O operations 7,000
Network Transfer Out 4,200
Total 3-year Cost 52,700
Cost per mosaic 1.46
Advertisements
This entry was posted in astroinformatics, Astronomy, Cloud computing, cyberinfrastructure, High performance computing, Parallelization, Uncategorized and tagged , , , , , , , . Bookmark the permalink.

5 Responses to The Cost of Running A Service on the Cloud Compared With Running It Locally

  1. Hi Bruce, this is a very interesting comparison. However, as I think I had asked you at the AAS or ADASS, one also needs to add the cost of personnel maintaining the local storage and computational equipment. I believe that once you factor this in, cloud will look much more attractive, if it’s not already.
    Great work.

  2. astrocompute says:

    Hi Alberto. Thanks for your comment. Actually, the local costs include estimates for such maintenance – they are absorbed in the numbers rather than called out as I am not permitted to reveal these figures. Cheers!

  3. Hi Bruce,

    ah yeah, the costs no one can disclose… 😦

  4. Steve B says:

    Thanks for posting these Bruce, the figures and commentary are quite informative. It’s clear that even cost reductions in cloud storage and network transfer rates can still prohibit large file transfers, like those for Montage. I look forward to your future post about CPU-bound computations.

    One comment: has IPAC ever looked into a distributed computing project like BOINC for any of its data processing needs?

    • astrocompute says:

      Thanks for the reply. The compute-intensive post will be in a couple of weeks, as there are one or two other posts I want to make first. Thanks for the link to BOINC. I will have a look at it. To my knowledge, IPAC has not looked at it.

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