followup to another conversation

For reasons I don’t quite understand WP seems to have eaten Dan’s post.?? Here it is, re-replicated:
Looks like we both have trouble being succinct!
Thanks for the Linux lessons.?? I’ve worked in both Windows and UNIX environments professionally, but I’ve never done any real *work* with Linux.?? I appreciate learning more, and your information is helpful.
There’s no doubt that you are 100% correct about one thing:?? the market will decide.?? I don’t know where WCC will end up, but I’m sure it won’t be a Linux killer.
One aside (and I’m not trying to throw fuel on the fire, nor am I trying to play Crazy Windows Advocate)–did you know that both Windows and UNIX still outsell Linux in the server market??? I read that recently and was quite surprised.?? http://news.com.com/Windows+bumps+Unix+as+top+server+OS/2100-1016_3-6041804.html.
Linux’s growth is explosive, of course.?? Anyway, as I said before, I’m not trying to tell people to choose a particular platform.
Thanks for the discussion.

7 thoughts on “followup to another conversation”

1. Well, I hope I am 100% correct about more than one thing.
In terms of sales figures, we have heard this many times before. Windows has been a larger market than Linux for a while. Will be, for a while.
Here is what is interesting about these numbers, and please bear with me on this. During the year that they are reported, the following information presents itself from the data:
1) The overall server market, dominated by windows and unix at 35B$US, grew at 4.5%+/- a little. 2) During this same time frame, Linux servers grew from 4.3B$ to 5.3B$US. Quick back of the envelope calcuation here. 23.3% growth in the Linux server market. This means that Linux servers as a market is growing at something like 5x the rate of growth of the overall market. Reasonable questions exist in terms of what is running on the blade servers, and those 6.8M low end servers that were shipped. I agree that the data is interesting. I think what we are seeing in the market is a rotation away from the standard solutions of the past and into new solutions. HPC doesn’t represent much of this market, a small portion in reality. Most of the Linux growth is outside of HPC by volume, though due to the rate of growth of the Linux cluster market, larger fractions of the Linux server market may be seen in terms of HPC. And this is the rub. If Linux were as hard, or as painful as the marketing implied, this growth wouldn’t be happening. If it were as expensive as is implied, this growth wouldn’t be happening. On the other hand, if Linux based systems had proven to reduce costs and increase productivity, the market numbers should reflect this with strong growth across the board in all areas. Curiously, this is exactly what is seen. My point about this is that when we talk about the market will decide, it is important to note that it has, and we see the impact of the decisions in the data. If Vista comes along, is the best thing since sliced bread, very easy to use, install , support, low cost to deploy, flexible beyond belief, well, it may change the equation. In the HPC market, and now spreading to other markets, what has occured has been this repeating cycle of a new technology that posesses 80% of the current functionality, for 20% of the cost. Each new technology drives down the cost, and this drives the volumes. Whats really interesting about this is that “traditional” Linux server vendors are breaking out into the non-HPC markets. They aren’t the only ones either. 2. Ok… thats$35B of a $49B market for Unix+Windows. Not a$35B market.
Note that blades are on a 100% growth path. They represent a cost reduction in terms of ease of management.
The mantra we see in HPC and in non-HPC is make it less expensive overall.

3. Granted, Linux’s growth is much larger than Windows. That’s one of the benefits from having a smaller piece of the pie. Windows growth is higher than 4.5%, however. I haven’t found any precise info, but the growth last year was double-digit.
FYI: I expect Linux to keep growing. I don’t think Microsoft is about to kill Linux by any stretch of the imagination.
As I said in a post today, I learned 2 interesting figures in this arena this week: 40% of Linux servers sold last year went into clusters. That jibes well with your point that Linux is branching out. The other is that Microsoft had 6% of the HPC market (no, I didn’t have time to jot down where these numbers came from). *That* number will increase over the next couple of years. How high? I don’t know yet.
I also heard a talk from Dr Richard Groves of Streamline Computing–they’re cluster builders from across the pond (they’ve built 7 Top 500 clusters, all *nix based so far). He thinks Microsoft will have a big impact in this area. His opinion is that having a single vendor providing an OS, the scheduler, the MPI Stack and the application API will make a cluster much easier to build, deploy, and maintain. They’re training their people up on it now.
That’s not to say that they’ll stop building Linux clusters by any stretch of the imagination. But they’re going to start building MS clusters.
License fees will continue to be an issue for Microsoft: people who won’t pay them certainly won’t buy from M$. Then again, many enterprises don’t run unsupported OSs anyway, so they can either pay support to someone like Red Hat or someone like Microsoft. License fees don’t deter everyone. Sorry to hear about 21st Century, by the way. That sucks. Better luck next time. 4. [takes technical hat off, puts business hat on] Actually the data on number of installed/sold servers with Linux is unfortunately incorrect. Linux doesn’t cost anything to install/copy/replicate. A company can take the same CD they use to install Redhat or SuSE and install this on multiple machines. They need support technically for a single machine if they desire support. There are no restrictions upon copying the binaries, there cannot be as they are under GPL. There are no restrictions upon installation (same reason). The only restrictions are upon support. Most businesses will happily pay for support, as it is a nice security blanket for them. But for a large server farm, they might only get a single copy of RH/SuSE, or use a site license. If they need support, they can buy it just in time. We have seen this at several customer sites now. What I am noting is what many others have noted: the linux installed base is undercounted. I have seen estimates all over the map, from a low end of 30% undercounted (30% more installed than counted) which I don’t believe, to 1/7th counted and 6/7th uncounted (which I also don’t believe). The estimate I have from looking at customer sites is something closer to 2 installations (e.g. about 50% uncounted) at commercial sites, and much closer to 10% counted at acadmeic sites). That is, I think the growth rates for Linux are under-reported in part because the measurement philosophy simply counts units sold with machines, and does not count other installation methods. Another phenomenon I see more of (lots at academic sites, and more at commercial sites) is conversion. An administration group will be asked to provide more service for less money and will start converting some of their installed systems over to Linux. In doing so, they start aggregating more functionality at a lower cost (TCO and acquisition) to the system. I don’t have good estimates on this, but we see this uniformly. You don’t need CALs for Linux servers. Costs don’t scale with number of clients under Linux as they do under windows. I haven’t seen the 40% of Linux servers sold going into clusters number, and I would likely dispute that claim. The number of servers in clusters sold per year is relatively small (even accounting for 10-100x multiplying factor of number of servers within each cluster) as compared to the number of servers sold into the overall market. If 3000 clusters per year are sold, averaging 32 nodes each, you are talking about under 100,000 servers per year. Assuming your average server sells for about$5000 (roughly), your $51B total market would have about 10M servers per year. Of which clusters would be about 1%. You can monkey with the numbers one way or the other, but it won’t move that overall figure much. Now with this in mind, assume that the$5B linux portion of the market (this is the counted portion, excluding roughly 50% of the actual market), is represented in those servers. This is roughly, again, 10% of the servers sold. Or about 1M servers. The 100,000 servers in the clusters would represent something like 10% of the market.
These numbers seem to jive much more closely with reality than 40%. I don’t know who arrived at that figure, or how they did it. Even if you replace our average $5000 server price model with a distribution, or other artifice, again, it doesn’t change the analysis much. Linux has countably a sizable fraction of the server market. Why Microsoft chose to target the small fraction that is HPC is beyond me. And hopefully their analytical marketeers are tapping Mr Balmer on the shoulder pointing this out to him. I am not sure it is in Microsoft’s shareholders best interests to pursue markets this small in installed seats. Assume for the moment, that Microsoft captures the entire HPC Linux cluster market, which is fairly obviously what they are attempting to do. At$500/server, this is $50M in revenue. Even with services/upgrades, this would be under a$200M market. This assumes they completely decimate Linux in this market.
Again, it is fairly obvious that this is not a market Microsoft should be interested in from a financials viewpoint. Sort of like a VC investing in a small business to get it to \$50M. The IRR is terrible, and not worth their effort. Worse, there is a market leader already, with 98+% installed base, and they are growing like wildfire.
I hope for Microsoft shareholders sake that someone points this out to the senior leadership. I question whether the rationale behind their entry was a desire to grow with a rapidly growing market, or to try to outflank a competitor. The analysis strongly favors that outflank tactic, and lends very little support (the particulars of their entry methods) to a desire to grow with a rapidly growing market.
As I have said before, if our customers demand it, we will supply it. We haven’t seen the demand, or even interest, and we have asked.

5. As for the 21st century fund, we read the reviews, and some interesting features lept out, specifically that the reviewers thought they were looking at a scientific proposal and not a commercialization proposal. We thought it was interesting that they asked what scientific question was being answered by a commercial product, as compared to the well articulated and succinctly described commercial need which they agreed with. In fact they opened with “this product would be phenomenal”. Yes, we agree. So do 80% of the customers we have spoken with (across engineering, life science, scientific computing). After they pick their jaws up off the floor and roll their tongues back into their mouths, they ask when they can order it. I would like to give them an answer, but that answer requires capital that we don’t have yet.
In the mean time, finding people who are interested in getting products with huge markets into the market is a priority. So is growing our existing business. The two go hand in hand. If we have to build it organically, we will. I am reminded of this, and I hope in 10 years we can have a few more people fill out their anti-portfolio of missed opportunities.

6. Wow. Once again, thanks for the informative (and lengthy) response.
I’m not going to argue with the numbers; you make very good points.
-DC

7. … and I promised myself that I would be brief … heh…