More about the tactics: commoditized HPC coming from an MCSE/Best Buy near you

Microsoft is pushing its resellers to enter this market.


From here

“Partner readiness will be a key focus for Microsoft Canada,” Whittman said. We will be encouraging all Gold Certified Microsoft partners in the Advanced Infrastructure and Networking Infrastructure competencies to participate in an upcoming High Performance Computing “Airlift” to take place at Microsoft Canada’s Mississauga headquarters in early September. The focus of this training will include a deep dive into the product, the market opportunity, and sales support resources to drive adoption.

Uh huh. Remember those paper MCSEs? Clusters are far more complex to get right than a basic PC network. Diagnosing and solving performance problems on tightly coupled machines is non-trivial. This is going to be … interesting.

Viewed 15248 times by 3161 viewers

2 thoughts on “More about the tactics: commoditized HPC coming from an MCSE/Best Buy near you

  1. Joe – I’d be happy to share our partner training schedule with you and your readers so that you can attend the class and see first-hand whether we’re equipping MCSEs with enough knowledge to successful deploy, operate and manage a Windows-based cluster. The U.S. classes begin in late July and run through August, and are one day long. Let me know if you’re interested.

    Patrick O’Rourke

  2. My post above was somewhat tongue-in-cheek, with a sardonic character to it.

    Feel free to post this. Basically my (and many others that I know) experiences in dealing with MCSEs are that these are generally people you want far away from your data center and important machines/information. Many took the courses to get the certifications, and as a result now seek higher paying jobs, without having the skills to do these jobs.

    With regards to having MCSEs deploy a windows cluster, I would be a little nervous about that for the aforementioned reasons.

    More importantly, as Doug Eadline points out over at Cluster Monkey, designing and building clusters and doing HPC is sadly not easy. The software load onto a cluster is to a degree, the second or third to final step in the process. It has to be done right, and this doesn’t mean in the context of the tools, it means in the context of the problem the cluster was designed to solve.

    Right now we are dealing with an organization with a cluster that was not designed. It was a pile of PCs. Stitched together with a gigabit fabric. I don’t want to make light of this situation. The customer is unable to use this to what they perceive to be its maximum capability. The maximum capability was “derived” by using what I consider to be invalid measures, specifically top-500 (HPL) runs. They have sadly discovered what I have long indicated, that HPL doesn’t represent most peoples workloads (which is why we created BBS specifically to enable people to encode their own workloads and create their own benchmarks). Moreover they are suffering from the impacts of the poor design. HPL by its nature encourages designs that scale in a particular manner, and don’t necessarily work the way the end user needs.

    My fear is fundamentally that the WCC product is going to encourage this “pile-o-pc” mentality. There is anecdotal evidence of the large PC vendors thinking of precisely this. Some of the bigger shops don’t get why something like 60% by dollar volume of clusters sold are not going through them. They figure if they can stitch together cheaper desktop units or other such things into a grid, then that will solve their problems and pull people away from the white box vendors.

    This means we are going to see more of the clusters that we are trying to help this customer run/use. They run into the design limitations very quickly, and it impacts their work.

    The point about it being available at Best Buy is one of the issues. If you go buy 256 desktops at $800/unit together, slap a simple network hub on it, run WCC, isn’t this a cluster? The answer is, no, it isn’t. This isnt a WCC issue, I see this with Linux as well. The problem is fundamentally good cluster design. Microsoft partners don’t really know how to do this. This is our forte. So is cleaning up after the messes that other “cluster vendors” make (sadly).

    At a different site from the above mentioned one, we have been dealing with the after-effects of an MCSE trained admin group. The results are not pretty.

    Post the training schedules if you would like, there may be interest.

Comments are closed.