The Free Edition of vSMP has zero value

Home Forums Using vSMP Foundation The Free Edition of vSMP has zero value

This topic contains 5 replies, has 5 voices, and was last updated by  Shai Fultheim 1 month, 2 weeks ago.

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
  • #7236


    Hi, Shai:

    I was pretty excited to find out that ScaleMP has a free edition for the community now. Unfortunately, it is my humble opinion that the free edition has zero value to society and please hear me out.

    Because of the fact that 16GB DDR3 quad rank ECC DIMMs are priced linearly now per GB relative to 4GB and 8GB DIMMs, 1TB memory limit on the free edition can be easily reached by just buying a Dell R910(with cheap X7550s) and fill the 64 DIMMs with 16GB ram($100 per dimm) instead of running a lesser server and run additional memory nodes with scaleMP.

    I am a researcher. I think an memory extension product of 1TB is useless. Because current software ecosystem already has in memory clustering solutions to alleviate DB or application memory capacity bottlenecks like Redis, Memcache, etc.

    In fact, I think the reverse situation is better, instead of giving the community a free memory extension product that can be easily replaced by simply buying a Dell R910, how about giving us the community a free limited CPU aggregation product? There are use cases(acadamic/research/analytics) that require only 16-32GB of ram, but can take a lot of CPU power. There are a lot of retired servers, particularly interesting is the Dell C6100 with Xeon 5400,5500,5600 series that businesses won’t use because it’s no longer the most power efficient. To researchers with a budget, those quad node servers could be aggregated by ScaleMP to get a 8 socket system, so there is social value to reuse those less power efficient servers for academic research purposes, instead of us buying Supermicro 5086B-TRF to get to 8 sockets(granted, the 5086 is only $10K, so it is not out of reach either)

    So if possible, I think a more potent “community edition” should center around a CPU aggregation product and put on a strict memory limit(32GB?64GB) to protect ScaleMP’s revenue line. I would rather see a 4 node Infiniband DirectConnect CPU cluster with a 64GB memory limit to allow academic old server reuse rather than a 8 node memory cluster with 1TB memory limit.

    Shai, think about it. Not only is it great advertising with University Professors/Students, it is also socially responsible for the acadamic environment to gain access to 8 socket Xeon L5520 servers(selling on ebay for as little as $50 now) It won’t impact your business revenue because no sensible business would tolerate the low memory footprint. But it is useful for academic researchers who do parallel programming.

    Thanks for your time and consideration.


    Shai Fultheim


    I appreciate your view and the lengthy analysis.

    You got it right – vSMP Foundation never intended to compete with off-the-shelf servers, so if one can get 8 socket system for $10K – he should do so.

    ScaleMP starts where the hardware ends! You should consider our products only where off-the-shelf hardware can’t deliver you what you need, and you are confined withing the boundaries of the HW.

    vSMP Foundation Free designed for customers with existing equipment (purchased over the last few years) that are looking to extend HW capability without acquiring some new boxes. It is allowing to grow without any hardware investment (or any investment at all), and serve as a good testbed for larger deployments.

    I hope this helps.



    Hi, Shai:
    Just to let you know that my post was never meant to low ball ScaleMP in anyways. I do see value in greater than 8 socket market for ScaleMP, and I was just poking you to see if you are interested in extracting value for aggregating quad dual socket machines to 8 sockets as a community experiment.But I guess that was never your intended market anyways. The question is, why NOT? There is a market for SGI Altix(or NumaConnect on IBM X3755M3) that ScaleMP will never be able to touch: those hardware solutions can run any OS other than Linux. For a software solution provider for virtualization aggregation, the marginal cost for producing one unit is literally zero. I personally think it is a great idea to arbitrage Intel’s 8 socket premium relative to 2 socket pricing. And a thread/memory limited edition is great marketing to community.

    BTW, according to the latest release notes, Xeon Phis require separate license. how is Xeon Phi licensed? Per PCI-e card as a socket or what? I think ScaleMP should be more transparent with Xeon Phi licensing as well as providing a 30-day Trial Version for potential customers to evaluate performance claims as well as sizing the physical hardware nodes. For example, I don’t know how many xeon Phis would be optimal for my workload (considering between 1-4 Xeon Phis hanging off of Dual E5-2600)

    Oh, last thing to say: you do have a great product. Sometimes, a revenue maximization strategy might work counter-intuitively. Most large companies with a deep vision tend to be more gracious with their community relations than small startups, and they usually end up ahead in the long run by sacrificing a little short term gain. I am sure you can understand what I am trying to get at. With the original post, I was merely trying to point out to you that a 1TB memory expansion product is not the best way to showcase ScaleMP’s value and that it is in fact cross-machine-boundary CPU aggregation that matters. For that reason, I think your pricing on the main page: $10K/TB for memory expansion and $400/socket for CPU aggregation is totally missing the mark. Because why would you do $10K/TB for memory when you can do Dell C6220 with quad node 256GB and only need to license $400*8sockets=$3200? Because you are overestimating market value for memory expansion and underestimating CPU aggregation relatively.

    Food for thought, really.

    Thanks for reading.



    Hello All!
    I am fully agree with TAO. And also want to add that there are serious doubts about the need to buy ScaleMP collection of processors without testing . Just not sure that the decision to assemble a Super Computer (VM) that must be based on the software ScaleMP that will to the satisfy projected needs for a specific tasks. I ‘d rather see ScaleMP Free edition with possibility to extend memory, CPU and disk, as well as a full fault tolerance for VM in case of any hardware failure. In my opinion the commercial version should be easy to extend for using only the limit by the number physical NODES. Commercical policy must be transparent, and the free version must demonstrate all the capabilities of the system only with limit for use physical computers. After all, buying expensive cat in the bag may be too risky business ….

    Many thanks in advance for good understanding.


    Shai Fultheim

    Hello Maxim and Tao,

    vSMP Foundation starts at $400 per socket, and allow you to set an VM of 2 nodes for <$2000. Subscription based pricing would put that at <$1000 - if that would make your cat too expensive, than I would suggest you look for other solutions. regards, Shai.



    +1 to what Tao said. It’s pretty evident ScaleMP doesn’t “get it”. They’re too busy being corporate interest tools.

Viewing 6 posts - 1 through 6 (of 6 total)

You must be logged in to reply to this topic.