Should SUN Push x86 More Aggressively to Catapult OpenSolaris?
I wrote a piece the other day stating that I had reconsidered my initial assessment of OpenSolaris, and that it may turn out to be a more powerful project than I had first thought. If you don’t feel like reading what I wrote, the bottom line is that OpenSolaris is developing a strong community and communities are powerful.
I think there is something else SUN could do, however, that would also have a strong impact on their ability to re-ignite the IT world’s interest in Solaris. Push harder into x86 and migrate away from UltraSPARC. I highly doubt SUN will do that, as they seem pretty bullish on the platform. Personally, I’m not so sure. Hear me out.
In the early days of SPARC, it was clear that a proprietary RISC architecture was strategically very key for SUN. There was nothing from Intel at the time that could compete with the SPARC chips, and if you were in the market for a high end workstation or server then a non-Intel platform was the obvious choice. Those days are long past. I’m not interested in the holy war of why RISC is so much better than the x86 architecture. I’m an IT professional, and what I care about in the real world is performance.
I used to work for a very large semiconductor company in Silicon Valley. When I started there, the environment was 99.9999% SUN (we had one HPUX box, as I recall, but that was for our software development group to do compatibility testing). Every desktop, every server was a SUN SPARC box. When I left that position, we still had a 100% SUN desktop environment, and our high end servers with lots of RAM were SUN, but we also had started building a somewhat large Intel/Linux server cluster on commodity hardware. The reasons may not be what you think.
While everyone is price conscious (or they should be), lower price hardware and the “awesomeness” of open source were not driving our move to Linux. Performance was. Take whatever position you want on the CPU architecture holy wars, the bottom line for our engineers was that their applications would run faster on 2 GHz Intel boxes than they would on 900 MHz UltraSPARC III boxes. And the performance for us scaled linearly with the clockspeed. A 2 GHz Intel box would run the engineering jobs approximately twice as fast. As with all things, your mileage may vary depending on the types of computation you’re doing so before you flame me, benchmark your own apps and see if it’s true or not. For our IT group, as soon as the application vendors were releasing Linux binaries, our engineers were eager to run them on Intel hardware.
I don’t critisize SUN for not having the clock speed of the Intel and AMD chips. Intel and AMD are large companies with a tight focus on the semiconductor business. SUN is a systems vendor, and its semiconductor operations are only a part of the overall business. Hardware engineering, especially for something as complex as a general purpose CPU, is incredibly capital intensive these days. I don’t think that it makes much sense for SUN to compete with Intel and AMD on their own turf. Why not just stick with a commodity general processor, and innovate around system design of scalability, reliability, servicability, and software?
One of the great marketing mantras from SUN has always been that they had a single architecture, and single OS. You could run the same apps on their high end servers (for example the SunFire line) as you could on a SunBlade 150 desktop. SUN ads and press releases would contrast that with IBM and their array of processors, system architectures, and operating systems. While it’s clearly a marketing campaign, I think it’s a compelling story. Unfortunately, SUN is moving away from it.
Today SUN has their UltraSPARC IV line, and their AMD line. The AMD servers are confined to the bottom tier of server offerings, and of course there is no binary compatibility between applications running on UltraSPARC SunFire servers and AMD Linux (or even OpenSolaris) desktops.
If I’m a prospective OpenSolaris “switcher” (to steal a word from Apple’s marketing department), I’m very likely going to be looking at OpenSolaris on an Intel or AMD system. Right now, that means I can run OpenSolaris on generic PC hardware, or one of SUN’s low end AMD offerings. It would be a lot more compelling to know that I could also scale up to something like a SunFire server with no code changes and no recompiles.
In my previous employment at that semiconductor company, we would have probably been much more interested in OpenSolaris than Linux if it had been around. Unfortunately at the time, SUN had just announced they were abandoning the Solaris x86 product all together, and then they announced that no, actually they weren’t. Not very reassuring for someone looking for an Intel-based UNIX solution.
It didn’t matter, though, because all of our third party software vendors were releasing Linux platform products, not Solaris x86 binaries. SUN is going to have to work with ISVs to get their support of OpenSolaris/Solaris x86. A Linux binary compatibility layer is not going to cut it for enterprise software. Even if the binary compatibility layer works perfectly, there’s no way for an IT department to justify why they are running millions of dollars of software, that the vendor supports on Linux and not Solaris, on Solaris systems under Linux compatilibity mode. The first time there is a problem with the software and the vendor’s tech support department won’t talk to you because you’re running this crazy binary compatability thing, executive management is not going to be happy. At all. ISVs need to support OpenSolaris.
I’m not so pessimistic about this, though. Just a few years ago, people were making the same arguments against Linux -the enterprise software companies will never support it, what happens when there are tech support problems or bugs, etc. It turns out that the vendors go where the market (customers) dictate. And that’s where OpenSolaris’ community is key- if the community is big enough, and loud enough, and vendors start losing sales, they’ll re-think their platform targeting.
From an IT perspective, it would have been a lot better for us at the semiconductor company to stick with a single operating system (Solaris) than to mix Solaris and Linux. Any time you add additional operating systems and system architectures into a production environment, you’re creating stress and additional requirements for IT staff. The IT staff I worked with were all comfortable with Solaris, but Linux presented new challenges, a new learning curve, new training requirements, etc.
While OpenSolaris may end up doing just fine as it currently is, I think SUN could give it a big boost and certainly spill a lot of ink in the trade press if they’d consolidate back to one CPU architecture- the x86 architecture.
Tags: OpenSolaris, Solaris, SUN, SPARC, x86
Related Articles:
http://blogs.sun.com/roller/page/jimgris?entry=rethinking_opensolaris