Again, coming back to this, adoption rates are critical. I appreciate Patrick’s point from post 306 that Microsoft doesn’t release this data.
In reading around Ken Farmer’s excellent www.winhpc.org site (sister to his excellent www.linuxhpc.org) site, I found a link to this article.
I recommend reading the whole thing. Especially page 2. Here is a quote.
The next portion of that paragraph is not something I understand.
Uh, sure. Not clear on what this means. Maybe they are ready for volume?
Further, the article quotes (the eminently quotable) Dave Turek of IBM.
And where we get to is a concluding sentence courtesy of Ms. Wu of IDC. Curiously I think it is right on the money, exactly in line with my business case analysis.
That’s gonna leave a mark.
They are not in this market for the profit. There must be another reason.
Sounds a great deal like what I had postulated. The business model just doesn’t work for them. Maybe it is a wealth redistribution mechanism, whereby the shareholders of Microsoft get to give money to partners. I hope not.
I just can’t get past the business model thing. As a small business guy, this bugs me. There won’t be primary billions of dollars in revenue. Hundreds of millions are doubtful at best. Tens of millions may be likely.
Ok, look at secondary effects. (Hypothetically) Lets seed a market, invest a few hundred million into promoting it, market the heck out of it. Pay cluster vendors to sell it. Will this drive other sales? Excel? Powerpoint? No, probably not. Well, not at additional hundreds of millions of dollars. Tens may even be a stretch.
Ok, what about tertiary effects. (Hypothetically) Lets seed a market, invest a few hundred million into promoting it, market the heck out of it. Pay cluster vendors to sell it. Will this drive other Microsoft technologies for consideration and usage, say application development systems. Again, app development systems as a revenue stream is kind of hard given the costs of visual studio and compilers. Add this to the barriers I keep mentioning.
So at the end of the day, I don’t get the business model. It doesn’t look like a traditional P&L; business unit. It looks and feels more like a tactical play. A play that hasn’t proven itself to date.
If there was a massive adoption, we would have heard about it in numbers. We haven’t. Ms. Wu appears to be confirming what I am (not) seeing.
Update: In that same article was this quote
Well, don’t know about “unix”, but …
The scary thing was the model I assumed in a previous post gave Microsoft about 1%. 2-3% isn’t bad, shows a huge error bar, which means that the accuracy of measurement is low, or the signal is not strong compared to the noise. My back of the envelope guess was actually pretty good if IDC’s numbers are to be believed (note: had a Gartner graphic from years ago showing how Itanium2 would be shipping 10’s of millions of units, and this is what convinced SGI to drop Beast and Alien a decade ago … wish I had a copy of that, shows how badly you can fool yourself and others with “analysis”)
Of the $10B HPC market, about $6B is in clusters. 10% of this would be $600M. 1% of this would be 60M. I estimated about 30M (0.5%). This is percentage by revenue. Lets use Microsoft’s own numbers. At $50k/cluster (of which about $10k is Microsoft’s), you need 20 clusters to hit $1M in gross revenue, $200k net for Microsoft. 1% of the market by revenue would be 20*60 or 1200 clusters! Woo hoo!!! That would be nice. Probably not real though. Because if there were 1200 Microsoft clusters a) Microsoft would be telling the world about it, and b) application developers would notice.
That last point is the kicker.
I am willing to bet that the IDC numbers are off, as the data is simply too close to the noise level to be measured with any meaningful accuracy (their noise signal is larger than their measurement signal). Unless of course, 1200 clusters have shipped. In which case I will be happy to withdraw my critique, and moreover, port apps over. 1200 clusters is a market in HPC.
As I pointed out in my analysis, I could believe 150 clusters. I have a hard time believing 1500.