Amazon New HPC Compute Cluster Instances. How much it costs?

I don't think from just reading the announcement, that many people realize the high costs involved in placing an HPC data center on AWS.

See  the blog from Jeff Bar at ,

"Each Cluster Compute Instance consists of a pair of quad-core Intel "Nehalem" X5570 processors with a total of 33.5 ECU (EC2 Compute Units), 23 GB of RAM, and 1690 GB of local instance storage, all for $1.60 per hour."

The price list for Compute Cluster Instance (CCI) is at . Please note the Reserved Compute cluster (not on demand) is priced $4,290 upfront for 1 year and 56 cents per hour, which apparently is about one third of the non-reserved , 6,590 for 3 year upfront and 56 cents per hour

Armed with information, let's price on demand, without commitment of time the configuration that ranks #146 on TOP500 described below:

" We ran the gold-standard High Performance Linpack benchmark on 880 Cluster Compute instances (7040 cores) and measured the overall performance at 41.82 TeraFLOPS using Intel's MPI (Message Passing Interface) and MKL (Math Kernel Library) libraries, along with their compiler suite. This result places us at position 146 on the Top500 list of supercomputers.

The [price for 880 CCI are then 1.60*24*30*880 = $ ! million per months, or 12 millions per year at 100% utilization. This quite mind boggling.

So if we want reserved instances, the same price will be 0.56*24*30*880 = $0.35M per months or $4.3 million per year., PLUS upfront fees for each new cluster added. For 880 clusters the upfront money to pay is 880*4,290= $$3.8 million. This results in 4.3+3.8 = $8.1 Million in reserved CCI

Sure $8.3M is lower than $12M by 31%. Note the "saving" - if we call these savings, are even higher if the commitment is for 3 years.

What does it mean

o Prices are easy to use
o Prices are designed to encourage long term commitments
o Prices very easy can evolve in high fees to be charged, even out of control
o Ease of use will democratize the HPC user access
o Many champion users will try to negotiate discounts,
o Profits are so attractive that other players will jump in

Over all good news. However to be able to use optimal scheduling policies in hybrid clouds HPC, billing is NOW a condition sine-qua-non. The comparison between the HPC on-site costs and the fees to pay Compute Cluster Instances, should be part from the cloud management software.

2 cents


Popular Posts