On Sun Microsystems Inc.
About AIX and Power
After last post about Mark Hurd preference of Linux and Opteron/Xeon over HP-UX and Itanium, I started to wonder if IBM had the same kind of low corporate commitment about its AIX operating system. Lucky me, it happened to be the case.
Lou Gerstner itself said that "...there are only two major growth platforms going forward: NT and Linux." . 34 days after that public statement he made a commitment for IBM to spend 1 U$S billion dollars developing Linux and no a single mention about expending even one dollar on the AIX operating system.
But corporations evolve and tend to fix passed errors (at least as soon as they detect their previous errors), which is a good sign, as no one can expect to always be right, but at least you can try to improve... That previous Linux endorsement by IBM was Lou Gerstner time, perhaps Mr. Palmisano have a different idea... Well it happens to be that Mr.Palmisano was one of the leaders in getting IBM to adopt the Linux operating environment (there is a good chance that he was responsible for Lou Gerstner Linux focus), so it is safe to assume that he will bet his company on the only one of the two “major growth platforms” (according to his mentor) that he doesn't have to pay royalties to use.
It was hard to believe that IBM would dump AIX in favor of Linux, but one CEO after another showed the same corporate direction... So I though that inside IBM there should be some kind of guerrilla against this movement. I went out and checked who was responsible for the AIX operating system. It turned out that Steve Mills is that person. Imagine my surprise when I realize that he was the most radical voice on the movement to replace AIX with Linux. He went so far as to state that Linux is AIX's logical successor... I wouldn't be happy with that statement if I'm an AIX customer... All I can say is that p Systems customers can expect to face tough times to get ISV to support Linux on the p Systems.
That's only the Operating System part of IBM's p Servers line. The other part is the hardware. IBM is really proud of their current Power5+ Processor and they should be. After all their p595 server is not that far behind from the Sun Fire Enterprise 25K server in performance (although they still lack enterprise class availability features like memory and CPU hot-swap, redundant system bus and fault tolerant clock-boards that do not requires a reboot). Times are changing, and now Power organization have just release a new open spec.
The whole idea behind having an open spec. is creating a bigger ecosystem (as SPARC have created) that will improve innovation. That was exactly what allowed new companies (ESA, Gaisler, Fujitsu, etc.) to create SPARC alternatives, thus allowing, Sun Microsystems Inc. to buy Afara, which brought to the company the great minds that helped develop our current and future CMT processors.
Thanks to SPARC being an open spec. Sun Microsystems Inc. is able to deliver the best processor roadmap in the industry, with a new system based on a new multithreaded processor in the next months, followed by an upgrade to the same systems which will then be followed by the amazing Rock processor for 2008 (if you want to read more about Scouting, one of the several Rock technologies, check out “High Performance Throughput Computing” Shailender Chaudry, Paul Caprioli, Sherman Yip, and Marc Tremblay IEEE Micro, May/June 2005, Volume 25 Number 3, p. 32). This is the most comprehensive processor roadmap in the industry, specially when compared to the Power organization roadmap (which clearly shows that IBM doesn't have anything planned after Power6) which IBM is promoting as its own Power Architecture roadmap.
Now IBM is starting to go along the same open spec. path with its Power organization. They say to have just release its first open spec. I'm quoting this as potential because I went to the site on Tuesday 21/11/2006 and register as a developer (the only free option) but haven't yet receive my user name and password to access the open spec (I guess the mainframe as web server is not fast enough nor such a good idea). Even now that they have post this open spec it is not clear to me if this spec is enough to create a binary compatible AIX capable system, or there is some other proprietary part of p Systems missing in the open spec. (if some reader of this blog can confirm or deny the possibility of creating an AIX capable system with this open spec I will fix this last statement to better reflect reality).
For customers to be of any relevance, this new open spec. movement from the Power organization have yet to deliver its first AIX capable sibling from a third company in order to start growing the AIX/Power ecosystem.
After years of being an open spec. (in fact SPARC is an IEEE 1754 standard), the SPARC processor went the last mile and the T1 implementation became the first mainstream 64-bits GPL processor in the world. What's even more amazing, its that just after nine months the OpenSparc site went live with the GPL Verilog code, the first T1 clone appeared. As I said in my first post of my blog (about Sun Microsystems Inc. long-term advantage), the future is all about communities, and Sun Microsystems Inc. is the most prolific community creator in the market.
If you are going to go and open your technologies to create a community around them, you better go and open the best of your technologies, so more people will be attracted to them. That's exactly what Sun have done with OpenSolaris and OpenSparc communities. Take for example a single T1 chip, you will need more than three and a half Power5+ dual-core chips at 1.9GHz to deliver that same performance. Assuming that Power6 will double the current Power5+ performance, it will still be behind the T1 processor (which shipped in 2005), and a lot more behind the T2 processor planned for late 2007 which will double T1 performance.
Customers should ask IBM what are the expected performance gains of the Power6 processor compared to the Power5+ processor, and what will happened after Power6, since it doesn't seems to hold a chance against the new CMT processors for Sun, “Project Olympus” (early 2007) and Rock (mid 2008) which will more than double and more than quadruple current UltraSPARC-IV+ performance respectively (according to David Yen's roadmap presentation).
As the great Erik Fischer usually says, “everything is invented at least twice in the microprocessor industry”. Take for example the multithreaded processors... the first implementation was the CDC6600 periphery processor.... Seymour Cray tends to be every were you look... What's even more impressive is that IBM is stating that they pioneered multicore, when Power4 was released in 2001, and the MAJC processor from Sun Microsystems Inc. was released in 1999... perhaps the meant the other kind of “pioneered”. Clearly MAJC was well ahead of its time (Rambus memory being one of its problems) and as far as I know Sun Microsystems Inc. only used it in the XVR-1000 and XVR-4000 graphics boards. It still makes a really good read if you want to know a little more about this amazing technology (agnostic pipelines and more).
Sun Microsystems Inc. have the “Microsystems” part as a very important strategy for the company. We are not about Operating Systems, we are not about processors, we are about Systems. That's why we not only exceed in our SPARC architecture but we also exceed in our Solaris Operating System. Now that Solaris became Open Source, our ability to innovate have increase several times. Just take a look at all the future technologies that Solaris will get from the Open Solaris community. Besides the amazing Dtrace technology, Solaris will keep evolving at a faster pace than all other Operating Systems since it is the only to have both, an incredible advanced community and a top 500 corporation on top, both innovating on its code. Add this to a growing SPARC community that is creating the fastest chips ever, and it becomes obvious why Hewlett Packard and IBM are moving towards Linux, it is the only community left they can try to leverage. They have failed to create communities around their own operating systems (in fact IBM is strong on its close source position for AIX) and are lagging behind the SPARC community on the processor flank, they can only pray for Intel and AMD to help them staying competitive in this last arena.
See you next post.
Posted at 04:37AM Nov 22, 2006 by maitas in Sun | Comments[3]
Acrobat Reader 7 on Solaris x64
Apparently it is possible to run Acrobat Reader 7 on Solaris x64 with BrandZ, at least this is what these guys are showing in their screen shot.
By the way, you should really give it a try to the latest Nexenta Release. The only problem is that the LiveCD it is not yet available. Nevertheless it is a really good product.
See you next post.
Posted at 11:16AM Nov 21, 2006 by maitas in Sun | Comments[1]
About Itanium and HP-UX
Oracle OpenWorld had several Keynote Speakers from October 22 to October 25. Mark Hurd (Hewlett-Packard CEO) was one of them. I always like to listen to what CEOs have to say, so I have clicked on the video, and was very surprised for what I saw there...
From the minute 21 to the minute 26 is located the most important part. He tells a story about hundreds of data marts that HP have internally that he will consolidate in one big Data Warehouse. To achieve this, he will use ProLiant and C-Class blades running Linux... no mention of Itanium or HP-UX whatsoever.
So I did a small research and found that HP runs SAP for its supply chain management. In fact, for year 2004 Q3 poor results they blame SAP for its performance.
That lead me to think that Mark Hurd would probably try to use the best price/performance system internally. Since he runs SAP, I went to the SAP benchmark site and found an interesting data, four Opteron 885 processors deliver 8520 SAPs and four Itanium Dual-Core 9050 (the most expensive version) deliver 7730 SAPs. What's more important is that both benchmarks use the same OS and DataBase (different binaries, of course), so the performance difference is only due to the processors and the system bus itself.
So now I understand why mister Hurd preferred to use ProLiant and C-class (Xeon or Opteron based, no Itanuim option whatsoever) with Linux (it seems that HP-UX licenses are too expensive even for him).
See you next post.
Posted at 10:27AM Nov 17, 2006 by maitas in Sun | Comments[3]