ProductionScale

I wanted to call out a specific post I just read. The reason is the because it highlights a very specific area of cloud computing that often seems little understood. That is, CPU Burst-ability.

http://www.thebitsource.com/2010/01/11/rackspace-cloud-servers-versus-amazon-ec2-performance-analysis/

I'll say that I've had this conversation dozens of times with various people and it's nice to see someone took the time to really do a good study. They really went all out to test under various conditions, times of day, different servers on the same platform, and compare apples to apples about as much as is possible. One of the sage things in the report is this gem, "Typically it is best to test specific applications to get a true measure of performance..." That is so very, very true for many reasons. I wish they had time to do the test on the Joyent platform as well. It's similar in it's CPU burst-ability but under the hood it's quite different. Info I provided about this was the subject of a couple of blog posts a while back here. For example, their platform uses 8 cores whereas Rackspacecloud uses 4 cores. Amazon, as far as I know, doesn't allow you to burst at all your CPU allocation is a hard cap which is quite different from RackspaceCloud, Slicehost, or Joyent.

Lastly, people forget and I keep saying ad nauseum, how you write the app means everything as to how well it will perform on the cloud. Generally speaking, adopting many of the ideas and processes around development of distributed applications is key.  But, this is also a very foreign concept to the bulk of developers I've worked with over the years.  This fact, how you write the application matters, is true pretty much no matter what cloud you deploy on and intimately tied to how well you automate certain operations tasks as well. Anyway, I know I've been a quite blogger lately. I simply have too much work to do to blog often with my startup. But, hopefully I can get back into the groove sometime soon.

This is rather interesting to see.  Someone already put up a set of live charts keeping track of the AWS compute instances.

http://cloudexchange.org

What's interesting to me is that the same resource can have different prices in different regions (obvious, but interesting) and that in many cases (if not all) the costs are substantially below the retail rate for the same instance.

For example:

us-east-1, c1.xlarge, $0.25 / hour.  The retail for that is $0.68 per hour.  Nice discount.

Now, for me, I just wonder when this page:
http://aws.amazon.com/ec2/#pricing

Will just become dynamic across the board.  The prices there just equaling the spot price.  What you pay for the duration you use the resource is the spot price.  If they did this they could just move to that model across the board.  No reserved instances.  No fixed pricing.  Just the market with price changing relative to supply and demand.  Seems like that would want to do this in time perhaps.

Very interesting.  If it was pure spot pricing then, in the event capacity became scarce in the context of demand then more money would flow to AWS as prices rise to help expand capacity.

At the moment a us-east-1, m1.small is $0.026 / hr.  This is still more that the $0.015 cost of a similar device in the Rackspace cloud.  So, to do a small video encoding job I'd still use Rackspace.  But, that can change pretty fast perhaps.

The one thing this spot pricing doesn't take into consideration is actual CPU resources available relative to price.

There was an analysis, I'm searching for the link will add in the comments if I find it, recently comparing different cloud servcies, their cost, and the amount of vCPU you get.  For example, if I say that the us-east-1 m1.small is $0.026 and the Rackspace equivalent is $0.015.  But, I get 4 vCPU's at Rackspace and only 1 vCPU at Amazon, then in reality Amazon is still far more expensive than Rackspace above and beyond the spot pricing vs. fixed pricing if you measure based on CPU Availability per dollar.

This is not a new problem.  It's one that is often quite overlooked when comparing the value of one cloud vs. another cloud.

My company, nScaled/LegalCloud.net has a job posting is on craigslist.  We recently added to our sales team and now need to add to the existing technology team.

Enterprise Cloud Computing Startup + Sr. Systems Administrator (sausalito)
http://sfbay.craigslist.org/nby/sad/1366310094.html

Cheers!b

Kent