Dr. Frank Soltis at COMMON: A Show Worth Watching
May 7, 2007 Alex Woodie
Going to see Dr. Frank Soltis is a rite of passage for System i devotees. Like nobody else in IBM‘s System i division, the chief scientist embodies the dual spirits of intelligence and independence that the Rochester, Minnesota, organization is known for. Last week at the COMMON conference in Anaheim, California, Soltis wowed the crowd again with his tales of past technological daring, his ill-conceived marketing plans, and promises of IT riches to come.
Soltis seemed cautiously upbeat as he opened his well-attended “Vision for System i” session in a room at the Anaheim Marriott, which is adjacent to the Anaheim Convention Center where the Spring 2007 COMMON Conference and Expo was based. “We’re doing a few things right finally,” he said.
He was referring to the new user-priced System i 515 and 525 servers that IBM announced in April. More than any other entry-level servers the System i division has put out in the last 10 years, the i5 515 and 525 machines finally come much closer to matching the cost of an X64 box running Windows, which have been cleaning the System i’s clock in the entry and midrange server market for years. (By midrange, we mean machines with four or fewer processor sockets.) The new boxes don’t knock Windows out the water on a straight price/performance basis, but they at least get the System i into the same ballpark, which is more than Rochester has been able to claim for years.
Soltis gave his opinion on what IBM should have called the new i5 boxes, if they had asked him–which it seems unlikely that they did. Soltis often jokes around with naming conventions for the box that he, more than any other IBMer, is credited with creating, as a way of playfully chiding Armonk and Somers about their frequent name changes. When the AS/400 was renamed the eServer iSeries, he says he would have just called IBM’s servers by first names, like Joe and Fred. Soltis, it seems, suffers with the rest of us concerning the name changes.
On this day, Soltis took a page from the AS/400 playbook and suggested that the new boxes be called the “Advanced System 36.” Apparently, Soltis sees parallels between the current market dynamics (lots of legacy boxes and a “new” platform with lagging sales) and those from two decades ago, when IBM had just launched the AS/400.
Back in 1988, IBM had many happy S/36 customers (the S/38 didn’t sell nearly as well and was more of a “cult” machine, Soltis says), and their reaction to being asked to move to the AS/400 was “no way,” Soltis says. “Over the years, there were a lot of S/36s that didn’t move. So we came out with the Advanced/36, which was the first RISC box,” he says.
The Advanced/36, which was really an AS/400 with a port of the S/36’s SSP operating system, made it so financially worthwhile for S/36 customers to move to the new RISC AS/400 servers that it became a huge success, Soltis says. To this day, there are still major international companies (who will remain nameless) that still run their applications in i5/OS’ System/36 environment, Soltis says. “And they won’t touch it because it continues to work.”
Similarly, the increase in price/performance of the new System i 515 and 525 boxes are so great that they could be responsible for the resurgence of the System i market. Only time will tell if the new boxes are attractive enough to get people to upgrade their aging AS/400 (pre-2000) and iSeries (2000-to-2005) boxes and kick-start sales for of System i servers, but it’s safe to say the Advanced 36 name won’t be trotted out of the closet any time soon.
Soltis also played the name game with the upcoming servers based on the new Power6 processor, which he says will ship later this year on other platforms and come to the System i line in 2008. Perhaps they will be called System i models 615 and 625, Soltis says. i6/OS anybody? Maybe IBM will go back to the past in the future again and rename i5/OS as CPF or SSP, the names of the S/38 and S/36 “operating environments.” (IBM was hesitant to call them operating systems back then and potentially scare customers away). Nobody really knows what it will be called, and, frankly, it’s too early to speculate. “Just when you think there’s a pattern, somebody will change it,” Soltis said.
What is worth speculating a tad about are the incredible new capabilities that IBM is developing with its Power6 processors and its follow-on, the Power7 processor, which is on tap for 2010, according to Soltis. “We’re on a path to create a system that can run anything,” he told the audience. One box will be able to run any operating system and any platform. “Fundamentally, that’s what we’re doing, that’s where we’re going,” he says.
But of course it’s not that simple. While a great deal of virtualization and technological flexibility has been built into the System i, the mainframe isn’t nearly so adept at adapting to new paradigms, the good doctor said. That is largely due to the great deal of assembler code still in use, and the different processors used in the System z mainframe. IBM plans to overcome this hurdle to integration by developing a pluggable architecture that will allow System z processors to connect directly to Power6-based servers, thereby allowing mainframe workloads to run side-by-side with Power6 operating systems like i5/OS, AIX, and Linux. With Power7, it is expected, the integration hurdles will have been overcome, and the architectures will be united on a common platform, Soltis says.
Similarly, Power7’s pluggable architecture will also allow X64 processors to plug directly into Power7 machines, Soltis says. While Soltis didn’t say it, this is the result of IBM adopting X64 standards with the Power7, and building the sockets for Power7 processors to the same dimensions and as X64 processors. BladeCenter systems will “plug into System i. That’s where were going from a hardware perspective. It’s all the same,” Soltis says. While nobody is saying Windows will run on the Power processor, the two entities are definitely growing closer together. IBM and Microsoft had discussions on running the next version of Windows Server, codenamed “Longhorn,” on Power, according to Soltis. And who knows–we might see something yet, although Soltis says not to get your hopes up.
Microsoft–the most successful software company in the world, and a one-time AS/400 user (although rumors persist that it still uses the servers)–envies the System i’s integrated architecture. Soltis trotted out an old quote of Bill Gates saying that Windows will “someday run like the AS/400.” Soltis said Microsoft was almost there with the Windows File System (WinFS) that had been planned for Longhorn, only to be cut out in 2005 to hasten development and get product to market, and related memory and security enhancements planned for Longhorn. “If you read [Longhorn’s technical documentation], what they were developing was Single Level Storage,” Soltis said, referring to one of the hallmarks of the i5/OS architecture. Without WinFS, there just isn’t much to Longhorn (or Longway, as Soltis jokingly referred to it). “So they didn’t really accomplish much,” Soltis says. “It’s amazing it took them five years to draw a few icons on Windows XP.”
Soltis also took shots at Intel and its marketing of dual-core processors as being twice as powerful as single-core processors. The dirty little secret, Soltis says, is most applications don’t run twice as fast on dual-core processors. In fact, most applications actually run slower on dual-core processors, and the reason behind that is most applications were designed to run in a single processor thread, so adding extra cores doesn’t do anything to boost performance. They run slower because chip makers compromise on processor speed when developing dual-core processors. Instead of selling a single-core processor that runs at 3.0 GHz, for example, it will combine two slower cores running at 2.6 GHz each, which theoretically should yield a boost in throughput. But because the application is only running on one of those (slower) cores while the other one sits idle, it actually slows the application, Soltis says.
But there is an easy fix. “All you gotta do is rewrite your applications,” Soltis says. Unfortunately, it takes three times longer to write and test a multi-threaded application than it does to write a single-threaded application, due to the nature of non-determinacy and the fact that all possible data combinations need to be tested, Soltis says. The System i is able to skirt around this issue not because most i5/OS applications were written for multi-threading, but because IBM figured out a way to run system workloads on the second thread, thereby opening the first thread for more work. Soltis says he gets razzed by a fellow computer science and engineering professor (Soltis teaches at the University of Minnesota) for recommending threaded applications. “It’s the worst thing to teach,” Soltis says the prof tells him.
To hear Soltis tell it, multithreading is a huge scam perpetuated by server makers and their marketing departments on an unsuspecting populace. “Intel dumped HyperThreading because nobody rewrote their applications,” he says. “Every single vendor that has done multithreaded processors has gone out of business.” The latest vendor to tout multithreading is Sun Microsystems. “I’m not saying sell all your Sun stock. I sold mine, but don’t quote me on that.”