Faith is the evidence of things not seen.

Therefore, under this definition, we must believe Amazon offering of a configuration made up of actual physical processor instances, not virtualized ones, that gets a #146 ranking on TOP500 for a cool $8M to $12M per year.

Quote:

“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.”

“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.”

Are we already in the HPC on demand era? I contend that yes, we are getting there but that there are some details to fix. Robert Jenkins, co-founder CloudSigma comments on the Cluster Compute instance:

“The biggest bottleneck for high performance computing is disk performance…. Adding 10Gbps networking will help, but will not solve the constraint of having remote SAN storage from the computing. Infiniband is much lower latency than 10 Gbps networking if this really is your bottleneck.”

For now, the Amazon Web Services Compute Cluster Instances the offer is restricted geographically to the Northeastern US and Virginia. CloudSigma offers a similar configuration to customers in Europe.

What is interesting here is that Amazon is a global behemoth company and CloudSigma is a recent startup. They are not unique, yet it is worth mentioning. For now they don’t even compete, as their services are in different geographies. One day they will, and it is nice to see that HPC customers have alternatives.

Chris Dagdijian from Bio-Team – the company that offers professional services to implement for their clients in bio-science and financial services HPC applications on Amazon, has this comment the AWS Compute Cluster Interface:

“Performance of the root/boot disk is way slower than any other type of block based storage. This is to be expected as the boot disk (even though it comes via an EBS-resident AMI) does not get the benefit of paravirtualization acceleration. The take home message is that the boot/root disk volume should not really be used for anything. This also means that this blog post showing how to increase the size of the local OS disk is useful only for playing around and not for anything serious.”

It seems the solution to slow boot disk access is adding para-virtualization drivers and the pressure from customers like bio-team may implement these changes on Amazon. Meanwhile Bio-Team possible remedy is described here:

“Future BioTeam cluster building practices may use the ~800GB of ephemeral storage to service a NFS or parallel file-system that offers input data to pipelines running on EC2 compute farms. Since we can’t trust ephemeral storage for anything unique we’d have a second shared file-system (backed by EBS) to handle capturing pipeline results.

Obviously there is one other comparison to make — how do these performance numbers measure against the 1-disk. 4-disk and 8-disk EBS RAID0 stripe sets that we’ve been testing all week?”

The possibility of double shared file system makes the use of Compute Cluster Instances not trivial and any company adopting the HPC in demand will contract expert independent services. Bio-Team makes this statement to differentiate themselves from web and database developers

We are bioinformatics and HPC types specializing in life sciences, not people trying to build the next Twitter or Facebook. What we care about when it comes to AWS performance may not be what YOU care about. In particular there is a ton of Internet information concentrating on methods for speeding up random IO access patterns on AWS. In our work, however, we seem to be more bound by the speed of long sequential reads (and sometimes writes). Parallel and serial scientific/HPC computing is different from building giant websites or databases.

Note that CloudSigma are offering fast RAID 6 storage in their bundle.

The availability of options available for Compute Cluster Instances is a positive aspect of the offering. Having some hiccups is part of the early adopters game, and final results will pay. All asperities will be nicely taken care of and probably the prices will come down.

What it counts in the end is how many customers use the solution implemented in AWS, Cloud Sigma and others. Allow me to paraphrase David Chernicoff blog from Zdnet:

“For the long term, everybody on the backend playing nice will be much more beneficial to cloud consumers than the same type of religious war that tends to crop up anytime you get Mac, Windows, and Linux users together in the same place. As much as zealots like to think that their personal opinion really matters, most of the business world just doesn’t care what operating system or application is being run, as long as it aids them in getting their work done and doesn’t impede them.

So whether the cloud application that is driving the business forward is running on OpenStack, Azure, VMware, RedHat Enterprise, or CP/M isn’t an issue to consumers of cloud services. Results are all that matter.”

This is the final reward for those – basically all readers of HPC Cloud – who had faith in HPC on demand.

Miha Ahronovitz is the Principal of Ahrono Associates. The company addresses the issues that 90% of the corporate software products fail to reach the initial goals. Ahrono specializes in cloud software products and business models. It manages the process to deliver products that users and clients really want. Miha was managing the Product Strategy and Business Development, HPC Grid and Cloud Computing at Sun Microsystems. He was also the executive product manager for Sun Grid Engine, now re-named Oracle Grid Engine. After Sun merger with Oracle in February 2010. he founded Ahrono Associates with two other Sun seasoned partners: Gregory Shirin and Kuldip Pabla.

Miha is also one of founders of Gridware, the start-up company acquired by Sun in 2000. Gridware flagship product was Codine, that later became, Sun Grid Engine. Miha’s blog, The Memories of a Product Manager is popular in the cloud computing community. His blogs on “Making big money with Hadoop” and “New HPC Compute cluster Instances” have resonated well with HPC cloud-oriented readers.