Bigger and faster AS/400s are coming. They will be as fast as we had been led to believe by IBMers in Rochester. But the more powerful 9406 boxes are not going to be available as quickly as many high end AS/400 customers might like. IBM Corp’s most recent road map for AS/400 processor technology, which is, as always, subject to change, shows that in September, IBM will announce enhanced 9406 RISC systems (CI No 2,926). They will use a faster version of the A30 chip set – running at 182MHz instead of the current 154MHz in the 530s – as well as using V3.7 of OS/400, which has been tweaked to make better use of IBM’s RISC processors. The new chip set, which Rochester insiders call Deep Sort Muskie, is a version of the Muskie processor that can run at the higher speed without any problems. The biggest system using this Deep Sort Muskie chip set will still only be a four-way machine, however. We’d been led to believe it might be possible for IBM to have an eight-way 9406 on the market by year-end, but this is appa rently not the case. The improved four-way model due in September will have about 182 RAMP-C relative performance units of power, which is about 53% more powerful than the current four-way 530. A significant portion of the performance improvement promised by the Muskie kicker will come from IBM’s tuning of OS/400 and related systems functions.
A 10% power bump
Customers that have 5XX systems will also be encouraged by IBM to upgrade their system software once again. But the odds are that the pain of jumping from V3.6 to V3.7 will be worth the trouble. IBM says that users with four-way 530 systems could see as much as a 30% boost in throughout from the V3.7 upgrade. Those with two-way systems will only see 15% to 20% improvement. Customers with uniprocessors – 4XX as well as 5XX models – will probably only see a 10% power bump. The enhancements coming in V3.7 are aimed at customers that have their processors running at close to full capacity and who are encountering software bottlenecks in the operating system and database management system that are related to high-throughput on-line transaction processing. Customers using a lot of database functions as well as journaling and commitment control will see the most gains from moving to V3.7. Customers that use plain RPG applications for on- line work and have heavy batch components will see only minimal improvements using V3.7 compared to V3.6. IBM will not have truly different AS/400 processors until the third quarter of 1997. That’s when IBM will announce 9406 machines that use the Apache chip, which is a single chip implementation of the PowerPC AS microprocessor. Some of the speed of the Apache comes simply from the smaller size of the chip. The current A30 chip set has seven separate CMOS chips connected to each other using fast bipolar electronics. Getting signals from one chip to the others in the A30 set takes time. The bipolar links also generate a lot of heat, which limits the speed at which the chip set can run. Even if IBM just shrunk the A30 circuitry down to one CMOS chip from its current seven, the resulting chip could run at a considerably higher clock speed. As it turns out, the Apaches won’t just be shrunken A30s. They will also be improved chips, so they won’t even have to run at the clock speeds of the fastest Muskie processors. The fastest Apache chip, in fact, will run at 125MHz, which is the speed of the slowest Muskie. But this top-end Apache uniprocessor will be nearly three times as powerful as the fastest complex instruction set processors used in the Advanced System 320 group and nearly twice as fast as the 154MHz Muskie.
The Apaches, which will support AS/400 as well as Unix memory addressing schemes, will also be used in IBM’s RS/6000 line of workstations and servers. The biggest 9406 that IBM expects to announce in 1997 will use eight of the Apache engines. That eight-way system, at 509 RAMP-C relative performance units, will have about five times as much power as the current four-way 530 RISC box. In 1998, IBM will still be using the Apache chips, but the high-end machine will be a 12-way box instead of an eight- way. IBM’s current plan does not call for an increase in the Apache’s clock speed, but we wouldn’t be surprised if IBM has a plan in the works to be able to speed up the Apache chips should its more advanced PowerPC chip design projects go off their schedules in late 1998. The Apache systems will require Version 4 of OS/400, which will include the microcode enhancements that are necessary to support N-way systems that have more than four processors. (IBM has said in past interviews that this piece of code will, in fact, be included in the V3.7 release that is due in September). IBM has not yet said what else will be included in OS/400 Version 4, but the rumor on the street is that V4 contains a true Unix kernel, which will make IBM’s software development efforts – e specially in the Internet-intranet areas – a whole lot easier. In the third quarter of 1999, IBM will deliver a super PowerPC AS chip, one that will likely use elements of IBM’s forthcoming PowerPC 630 chips as well as other advanced PowerPC technology. This chip, code-named Northstar, will not simply be a faster version of the Apache chip. It will be designed from the ground up as a unique member of the PowerPC family of microprocessors. It is unclear whether or not IBM will use Northstar in RS/6000s as well as AS/400s, but if IBM’s PowerPC 620 and 630 efforts don’t pan out, it is likely that IBM will use Northstar in its Unix systems as well as AS/400s. This saves time as well as development money, both important to IBM right now. The fastest Northstar uniprocessors will run at 250MHz and have about 120 RAMP-C relative performance units of power, which means that the Northstar uniprocessor will have as much throughput as a whole four-way 53 0 system. A 12-way Northstar system will have about 978 Relative Performance Units of power. When you do the maths using IBM’s CICS/DB2 benchmarks, this big Northstar system will have about 20% more raw processing capacity than IBM’s largest water-cooled mainframe, the ES/9000 model 9021-9X2. This will be one very big AS/400. IBM is talking about the performance of the forthcoming Muskie, Apache and Northstar systems for a couple of different reasons. For one thing, high-end AS/400 customers get antsy about the future, especially those that are already on the fastest AS/400s in the line. They want a clear idea of what’s coming down the pike so they can do their capacity planning. IBM is also talking about future AS/400 processors because Intel Corp, Digital Equipment Corp, Hewlett-Packard Co and even IBM’s other divisions have all issued statements to the public that outline where they plan to take their processor technology and associated system software between now and the end of the century.
Customer needs
According to the IBMers we spoke to, the estimates of forthcoming AS/400 processor performance do not include IBM’s efforts to enhance AS/400 processors with Very Long Instruction Word technology, nor the use of OptiConnect or Non-Uniform Memory Access system clustering technology. Frank Soltis, the AS/400’s principal designer, has said IBM hopes to make more powerful processors using Very Long Instruction Word, and bigger AS/400 networks with NUMA, but IBM is apparently not depending on these projects to come to fruition in time to meet the data processing needs of its largest AS/400 customers. All that we know for sure about AS/400 software in 1999 is that the Northstars are expected to use a different release of V4 software. There is not, as yet, a plan to develop a whole new version of OS/400 for the Northstars. But if IBM’s processor development plans do not pan out as expected – three years is a long time in the computer business, so it is very hard to predict with any accuracy what semiconductor scientists and computer architects can accomplish – AS/400 customers may have to move to a very different operating system. This is especially true if IBM decides
it has to use either Long Word or Non-Uniform Memory technology (or both) to make bigger and faster AS/400s because it can’t get the Northstars ready for market on their current schedule. And between now and the millennium, the commercial data processing world could take some unexpected turns. Customer needs might change to such an extent that IBM has to modify or even scrap its current software plans, although we think that IBM’s core chip technology will progress more or less as outlined.
By Timothy Prickett
From The Four Hundred, May issue (C) 1996 Technology News of America Inc.