[Beowulf] MS HPC... Oh dear...
landman at scalableinformatics.com
Wed Jun 14 06:02:31 PDT 2006
Thomas H Dr Pierce wrote:
> Microsoft HPC will work IF the market wants the technology that they can
> supply. The answer is in market segments. And the Supercomputer market
I don't think the technology is what will win people over. Its the
> The Linux HPC market is on the market demand track from CDC -> Cray ->
> Vax -> Unix Workstation -> Linux . It has grown modestly, but not
> enormously. Dual CPUs have a bigger growth market and potentially the
I disagree, and I think IDC and a fair number of others do as well. The
market has grown in real $$ terms, significantly, during this timeline.
In large part due to the cost of supercomputing processing power
In the late 70s early 80s there were few users of supercomputing
systems, as there were few supercomputing systems to be had. Access was
limited, so usage was limited, and demand was limited. Cray measured
sales in 1 or at most 2 digits per year. The market size would be lucky
to be $100M.
In the late 80s to mid 90s, the smaller supermicros appeared, and
largely decimated the market for the supers. This wasn't their original
intent until much later (R8000 time period). In the process the market
grew about 10x to about $800M/year (old memory from historical data,
could be off). The user base grew much larger, mirroring the growth in
In the late 90s to mid 00's, the clusters began their ascension. Today
there aren't many supermicros left, many of the vendors are
gone/absorbed or in Chapter 11. A few are still pushing their wares.
In the process the market grew about 10x to over $9B/year (IDC data).
The user base grew much larger, mirroring the growth in the market..
This is a pretty good growth.
In each case a newer technology came up from underneath and "creatively
destroyed" the old market. It would not be wise to bet against it
happening again. And this market will again likely grow in dollar value
and user base by significant amounts.
> Technical issues will not maintain the market for cluster computing.
Of course not. End users insatiable demand for cycles will.
> Nor will technical best-of-breed win over uninterested markets. The
Ease of use and cost will.
> past is full of superior technical solutions losing to market accessible
> good-enough solutions ( eg betamax vs VHS and token ring vs Ethernet ..
> and the beat goes on...)
Agreed. The corpses of companies with "better ideas" and "better
products" in this market are quite evident. More to the point,
technological superiority does not guarantee success. A great marketing
organization can take inferior products and make them winners.
Conversely, great products can be destroyed by a terrible marketing
organization. The latter is the tale of SGI in the late 90s.
Whether or not Microsoft's HPC product is good technologically, they
have a great marketing group.
> So please desist discussing technical superiority. Technical issues are
> not the key issue in market segment growth of HPCs. So lets see who
> wants CPU power with windows integration and discover how big that
> market is.
I disagree with not talking about technical issues. It is one important
aspect. However, there is the Pareto rule that will come into play. I
am of a differing opinion as to whether or not this product matters to
the space as compared to others here. It does matter, and we have to
figure out how to deal with it. The "whether" to deal with it question
has been settled, it will be there.
All of that said, I note a number of things. The IDC and others data
suggests that Linux clustering has been wildly successful, growing at
gangbuster rates for more than 3 years that it has been on the IDC's
radar. It has done this without a marketing department of the caliber
that Microsoft can bring to bear. It has done this in very large part
by significantly decreasing the cost of processing, and being easier to
use and to port to than vendor *nixes.
Microsoft will be successful if they can also lower the cost of
processing, not over *nix, but over Linux. I don't think that is
likely. They would be successful if they can get commercial code ports
onto their platform and get exclusion of the Linux platform, in terms of
taking commercial cluster market share. Again, this is a highly price
sensitive regime, where if you ask a customer to spend another $8k per
cluster for the OS, they are likely going to balk at it (given the cost
of the cluster is in the $50-100k region, the OS cost would be in the
It remains to be seen if this will be successful in the market. It is
there however, and it would be unwise to ignore it or dismiss it out of
hand. And it is possible that it may even be a reasonable platform (no
I am not on Microsoft payroll, simply positing the "the elephant may
learn to sing" possibility). I won't know until I try it.
> Tom Pierce
> Bldg 7/ Rm 207D - Spring House, PA
> 215-641-7475 - Office
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax : +1 734 786 8452 or +1 866 888 3112
cell : +1 734 612 4615
More information about the Beowulf