Often times, things we say or write about are taken slightly (or massively) out of context, repackaged, and written or spoken about in a different manner that subsumes the original context or intent. Even well meaning people do this. It is all part of the process of forming an opinion, specifically an interpretation of events or speech. One might construe a specific person wrote something they did not.
Where am I going with this? Well, off to a SaaS blog. SaaS as you may or may not know, is the “new black”. It is the buzzword loved by entrepreneurs without a viable business model to sell to VCs who ought to know better. Think web 1.5-web 2.5. I am not bashing on SaaS, it has some merits. The economics of it simply do not (yet) work for customers despite protestations to the contrary. This is the ASP model of the mid/late 90s redux. But wait, its different, the servers are virtualized now. Yeah, that makes it all better. (how it solves the capital and upgrade problem of the SaaS hosters is another story, and an open and unanswered question unless you are Google).
SaaS *may* eventually arrive with a meaningful and potentially useful business model. Any model which relies upon customers paying based upon “value” rather than reducing their real costs is destined for failure. But that is true in the vast majority of markets. Including HPC.
And that brings about the connection. Way back in the early stages of this blog, I marveled at the lack of investment capital in HPC. In 2005 the market was $7B USD. In 2006 it appears to have been $9.2B USD. In 2007 it looks like it will be north of $11B USD. Overall market growth of over 20% CAGR per year sustained over about 1/2 of a decade so far. Demand is outpaces supply. Prices are not rising, and again, any business model requiring a significant price rise to purchase “value” is likely a failure-in-the-making. Value in the case of HPC is the delivery of more and more efficient processing cycles. HPC use is growing rapidly, it is branching out into many areas.
That is the context for my comment in the linked post. Unfortunately, the commenter lost the context. No one accuses eBay, Yahoo, Google, of doing HPC. They do something very very different.
Use cases and real applications for HPC are on the rise. They have been for a long while. Have a good long hard look at the Council on Competitiveness site. Some of the choice quotes that we have used in talking to investors have included:
A July 2004 white paper commissioned by the Council on Competitiveness (CoC) and conducted by the International Data Corporation (IDC) surveyed 33 chief technology and information officers from aerospace, automotive, life science, electronics, pharmaceutical, and software companies, to determine the HPC needs of U.S. industry. The survey results are compelling:
70% (of the CIO/CTOs) indicated their companies could not function without HPC. Over 25% could quantify HPC’s ROI to their business, whether it saved millions of dollars, shortened production development cycles, or provided faster product-to-market timing
These are from an article from Stanley Ahalt and Kathryn Kelley of OSC in (the now defunct) ClusterWorld Magazine in 2005. ClusterWorld morphed into Doug Eadline’s excellent ClusterMonkey site.
Lets be clear about this. HPC use is increasing, its users see tremendous value in it. More users are entering the market, the market is growing (it is not growing along an elastic fixed-dollar curve, as the cost to acquire and run drops, the number of market entrants is growing rapidly).
It was this I was lamenting, in that we have a large and growing market, with well defined needs, and very little new capital entering it. VC’s often complain about too much money chasing too few deals. Hmm…. One doesn’t have to think to hard to note that maybe the type and location of the deals they are chasing needs to change. This is sort of like the apochrypal story I heard attributed to TJ Watson (of IBM fame)
Two shoe salesmen found themselves in a rustic part of Africa. The first salesman wired back to his head office: ‘There is no prospect of sales. No one here wears shoes!’ The other salesman wired: ‘No one wears shoes here. We can dominate the market. Send all possible stock.’
or refactored for the context I am talking about
‘Two VCs found themselves without deals to close, one in Silicon Valley, the other out in the wider (fly-over) country. The first VC wired back to his head office: ‘There is no prospect of deals here in Silicon Valley. There are no more business plans, we have run out!’ The other VC wired: ‘There is no capital out here, lots of business plans, and great ideas and teams ready and waiting for capital, with low burn rates, low costs of doing business. We can have our pick of the best deals. ‘
Ok, back to what I was writing about. The point being is that real HPC users and consumers have very little to do with what the commenter noted. Again, cherry picking the site, he found my SC06 followup notes and latched on to point #4, without absorbing the relevant context. He surmised that I stated that:
He found that real applications and use cases for high-end HPC were missing this time.
Uh… no. I noted that this year at SC06 I saw few applications on the show floor. It was mostly a hardware show this year. A few brave souls showed off applications: these were mostly the research university and government lab sites. I don’t consider cluster Excel an application. If Excel were, I dunno, accelerated somehow, that would be a good thing. But it is the tool, and the thing that gets accelerated is the model within it. If this model is not open to parallel solution, a cluster enabled version of Excel will be meaningless.
I did not say, nor did I imply that applications and use cases for high-end HPC were missing. That my gentle reader is a misinterpretation at best. They were simply not seen on the show floor, and I lamented in the next portion of this paragraph that Maybe it was the venue, or the timing, or ??? I dunno. Applications are the reason that SC machines exist. That is while the tools are neat, its what you do with them that really matters.
As for the use cases. This was a cherry picking, repackaging, and re-targetting of point #4.
Here is point #4
Reasonable use cases and data for benchmarks. Here and numerous other locations I have asked users to give me reasonable use cases for storage and file servers. It always seems to degerenate into ???tell me how fast the disks are??? regardless of what the real use case is. This is odd as the only thing really worth asking is how your use case and other use cases you are really interested in perform. If you care about raw disk speed in corner case scenarios, sure, this might be interesting. But does it matter if your use cases never touch these corner cases? So a number of vendors put data out. We tried to compare our little JackRabbit against their offerings. While we are happy that our unit appears to out perform these other units in the test cases, we aren???t sure how this matters to the use cases. This may be due to the almost complete decoupling of use case from benchmark.
How this devolved into a missing use case for HPC is beyond me.
It was about benchmarking. I again lamented that most benchmarking has turned into a “show me speeds and feeds” versus “how well does it run the application.”
I consider that to be a problem. Machine comparisons are “natural” to do, but a single number which is strongly decoupled from the actual application performance with a real world use case is largely useless as a metric for comparing machines. A vector of numbers only weakly coupled to actual use cases are marginally better, but still not good. We have maintained for a long time that real world benchmarks are the only ones that matter in the end. Way back at SGI I remember customers telling me that there was simply no way that the pitiful 90 MHz R8000 processor could beat this 333 MHz Alpha unit on any test. It was impossible. Yet when we confronted them with real results from real benchmarks, and they had a hard time explaining how we won, they gradually were won over.
Single numbers of dubious value and weak to non-existent connection to your use case, are a poor basis for making decisions. This is the point we have been making. In the case of benchmarking our JackRabbit product, most users seem to want to get to the “all important” raw disk speed, despite the fact that their use case rarely gets there.
Thats what was worrying me, and this is why I wrote point #4.
HPC has great use cases. It has a large and growing market. It is hot in the customers eyes, just not in VC’s eyes. SaaS has uncertain at best use cases. It is a small market, maybe it will grow if someone can come up with a compelling “must have” app which is far less expensive to host than install. Right now most SaaS I have seen is in the “would be nice if” phase, and not the “must have, incredibly compelling and obvious” phase. And yet VCs are drawn to it, not unlike moths to a light. It is hot in their eyes. Not unlike the Web 2.0 (bubble). Or RoR. Or “grids” were. Or (substitute your own fad in here).
About the only thing sure of in the SaaS world is that lots of VCs are going to lose money. There are technological and business model risks. It is not a pure execution play. In HPC, it is largely an execution play unless you want to invest in new technology. Non-commodity computing technology won’t be viewed with great love in this market. Maybe this is what frightens VCs away. Cause you can’t make money in commodity computing markets and grow a huge business.
Someone ought to remind Michael Dell of this.