Canada Shared Service Vs. Amazon

I was weirdly excited when a colleague sent me the following article:

It describes a half-billion dollar contract between IBM and Shared Services Canada for a computer system to run weather and climate simulations. For years I have been reading Paul Ramsey’s blog about how government IT projects have used excessive resources while delivering severely tempered results. I have also been listening to my boss (who shall remain unnamed), talk about how all of his former colleauges in the Government of Canada have been hamstrung by the introduction of Shared Services, a top-down government IT service that dictates the solutions without understanding what are the actual requirements.

Shared Services has been blamed with causing the various government agencies to be unable to actually do anything with computers. On one hand, this has been good news for me. I can do something without being required to use Shared Serices. But on the other hand, I pay taxes in Canada and am frustrated by seeing our government pay hundreds of millions of dollars to do something that… I possibly could do, but am not being paid hundreds of millions of dollars for :)

So let me go back to this CBC article. IBM won a contract worth $430 million to build a new system for Environment Canada. That’s some serious money, which, I don’t necessarily have a problem with, if it’s actually delivering a reasonable value.

However, let’s do the 60 second, private market comparison here. Since the introduction of Amazon AWS, computational storage and processing have been considered, to a further degree, to be much like a utility. If you treat it like a utility, you can do a quick computation as to how much this would cost, if you simply purchased it on the open market. So let’s do the computation…

From the article and its links, the IBM system includes:

• 790 compute nodes at 36 processors / node
• 40 PB of disk storage
• 230 PB of “tape storage” (or “cold storage”)

If we look at AWS’s offerings, “disk storage” is approximately equal to Amazon S3 and “tape storage” is approximately equal to Amazon Glacier, both of which are priced at an exceedingly reasonable level. A 36 processor compute node is approximately equal to AWS’s m4.10xlarge node, which has 40 virtual CPUs and 160 MB of RAM.

According to the article, Environment Canada signed an 8.5 year deal for $430 million. If we assume that AWS’s current prices will remain in effect for the next 8.5 years (their prices have always decreased in price over the entire lifetime of their service), we could calculate the following price list:

• 790 nodes * $1.480 / hour * 24 hours * 365.24 days * 8.5 years = $87 million
• 40 PB * $0.021 / GB * 2 ** 20 GB * 12 months * 8.5 years = $90 million
• 230 PB * $0.004 / GB * 2 ** 20 GB * 12 months * 8.5 years = $98 million
• Total = $275 million

At first glance, it looks like the government of BC could purchase all of the required components of this system–archival backup, storage, and computation–on the open, commoditized market for around 64% of the price that they actually bought it for. Why? That is an excellent question…