Eric Raymond, wrote a short article on his blog, “Commoditization, not open source, killed Sun Microsystems”, which I commented on. This blog article says a little bit more than I felt I had room for on someone else’s blog, and I probably abused his hospitality there. I have thought long and hard about this, because I worked there and thought it i.e. the company was worth saving. Here’s what I said on Eric’s blog, and a bit more. I start by saying that the first thing about Sun’s failure is that it all depends on where you want to start; Sun’s failure was baked in long before the 2000 fall from profit.
Since the eventual winner in IT CPUs is Intel, perhaps Sun should not have doubled down on SPARC in the 1980’s. They had an Intel 386 SunOS workstation in the 1980’s and they licensed SPARC and critical software products (such as NFS) for peppercorn prices. The strategy was to create markets and then compete in them, sadly a bit too effectively since they competed with their licensees eventually becoming the only i.e. last surviving SVR4 UNIX vendor and ending up being one of only two SPARC vendors. However, the SPARC & other software licensing shows their early commitment to open source strategies and for many years they had an absolute microprocessor performance advantage.
Windows NT drove SunOS off the corporate desktop, partly because they didn’t fight. They’d started moving off the desktop towards the data halls by then. Linux ( & Windows ) enabled x86 CPUs to enter the Datacentre and compete with Sun’s Server offerings in their, by then home space. Sun moved up market, with the E10K which they bought and its successors and looked to monetise their intellectual property through hardware sales. They recruited a lot of marketing and sales from, the by now defunct DEC; people who only understood proprietary strategies.
They lost the Universities and their students to Linux for a number of reasons. The enclosure of Solaris made it unattractive to Universities since they couldn’t re-engineer (or even tune), you had to buy a compiler and it ran best on SPARC which they didn’t want for their own use, mainly for cost reasons.
Sun’s Linux experiments were just that. History shows that Sun should have built a desktop and a platform. It’s what customers see and developers use. Microsoft Office won the desktop battle because it got into the home and the desktop, and now the phone OS UI is why Apple are a successful UNIX company. Sun’s disruptive defensive strategies were pursued only half heartedly at best. They bought an intel port of UNIX and made it SunOS compliant and then SVR4 compliant, but they positioned at as competitive with the SPARC line and so lost the sales force and then announced its abandonment at Solaris 9. They were persuaded not to by six large customers; it would seem the damage was done. Announcing the non delivery of Solaris 9 on x86 was act of extreme stupidity and arrogance.
Eric asked in response to one of the commenter authors about Sun’s blades, whether they were too late; there were two generations of Blade server. The first was developed in the UK which of course was the first engineering team to be fired; the second were architected by Bechtolsheim and customised for specific HPC deals. In the latter case, I loved the fact that you could choose to make a chassis a storage grid or throughput/grid by selecting and using different, exchangeable I/O modules. Both generations of blade were a step back to revising what might have been Sun’s biggest mistake, shipping the early workstations with its network routing software disabled. Some argue that was the product gap that created CISCO. The big failure of these blades or PODs was that the sales teams wouldn’t sell them against the SPARC database machines. The first generation blades were killed in the corporates by the network i.e. backplane (software) architectures adopted, and in the Universities by cost. The number of people in Sun who spoke Grid could be counted on the fingers of one hand which was another reason the blades didn’t take off. Sun and its customers were wedded to the RDBMS, which was a scale up, not scale out solution and it was taken by surprise by both the HPC compute grids and the new age data grids as they became the future volume deployment paradigms.
The UK designed, pre Bechtolsheim’s return were beautiful, but underpowered and very noisy.
I am firmly on the side of those who argue it was commoditisation not open source that killed Sun. HP is a proof point, also showing that dreadful management isn’t necessarily fatal. Sun recognised the sedimentation and commoditisation tendencies but chose to try and stay in a high margin business; scale up hardware with a proprietary CPU.
Why didn’t Chip Multi Threading (i.e. Niagra based systems) take off? I am not sure, too long to get out the door, and by the time they did, Intel were in spitting range?
Long term readers here, will know I am also influenced by the arguments of Carlotta Pereza. I suspect she would argue that as IT enters its Deployment Phase, supply chains and markets mature and there is a squeeze on on super profits. This effect of the maturity of the industry is what leads to several of Raymond’s correspondents to state that you can’t make money from hardware. You can but the early phase super profits are harder to make. It’s just part of the evolution of the silicon age’s Kondratiev wave.
While the long term opportunism of Sun’s leadership didn’t help, I reckon open sourcing Solaris and other products staved off the inevitable, I can’t see a counter factual, if that’s the word we’re using, which suggests that a proprietary software strategy would have saved the company. Too many decisions that killed the company had already been taken. Innovation was happening elsewhere.