Although IBM Corp may not have enough confidence in RISC AS/400s to ship them in volume to customers, they apparently work well enough for benchmark testing. Late last year, IBM released its first TPC-C benchmark test results for the new machines. The TPC- C tests show the RISC boxes offer good value compared with previous generations of AS/400s. However, the test results also reveal that IBM, despite all its efforts, is falling behind upstart competitors, especially those selling souped-up personal computer servers rather than proprietary systems. If IBM doesn’t catch up in the mid-range, the company will be in trouble. It will force many AS/400 customers, especially those on lean budgets, to put new jobs on systems other than the AS/400s they now favour. The TPC-C tests are the most widely-accepted commercial benchmark ever developed. IBM has been using it to gauge the power of its AS/400s since the E generation. Starting with the RISC machines, IBM’s AS/400 Division has gone with a network configuration on the TPC-C tests instead of stand-alone systems as it has always used in the past. Component machines The network IBM ran in the tests had five AS/400s supporting the TPC-C workload. (TPC-C simulates the data processing associated with running a warehouse. It is in many ways similar to IBM’s RAMP-C test). At the centre of this AS/400 network is a 50S-2121. This is an Advanced Server, and it is the fastest of the single chip server models. It uses a 77MHz PowerPC 620 chip, called the Cobra 4, that has been stripped of its symmetric multiprocessing features. The machine has been equipped with 1Mb of Level 2 cache memory to speed it up. The 50S has been configured with 1,024Mb of main memory and 80Gb of disk capacity. Four 400-2130 Advanced Systems are connected to the 50S by two Token Ring local networks. Each 400 model is configured with 128Mb of memory and 2Gb of disk. The model 400s all use the same Cobra 4 chip as the 50S does, but they run at only 50MHz and do not include the L2 cache memory. All of the AS/400s in the network are equipped with OS/400 v3.6. Going by the TPC-C tests, this network of AS/400s can do about as much work as an F90 and slightly more than a 320- 2051 or an E95. The RISC AS/400 network does a lot more work than you might expect, mainly because each of its component machines has been configured with great gobs of memory. The 50S is at maximum memory, and the 400s are at 80% of their maximum. The E and F systems used in IBM’s first two years of TPC-C testing were configured with modest amounts of main memory because AS/400 memory was much more expensive then ($765 per Megabyte) than it is now ($85 per Megabyte for the deskside boxes). The performance gained by adding extra memory didn’t offset the added cost; price-performance would have suffered. The TPC-C tests point out that IBM has fallen behind others in the mid-range in two ways. First, the RISC AS/400 chips don’t pack the punch many had expected based on the raw performance figures for PowerPCs. By Timothy Prickett IBM may have met its extremely conservative RISC AS/400 performance estimates, but there is little question that IBM had hoped far to exceed these levels. Second, IBM’s hardware and software costs too much compared to other mid-range systems. Judging from the SPEC series of chip tests, it should take six PowerPC 601 chips (at 75MHz) to do the work of four Pentium chips (at 133MHz). IBM’s TPC-C tests with RS/6000s back up this ratio. Now let’s move ahead one generation. According to SPEC, the PowerPC 620 running at 154MHz should have about 25% more oomph than a Pentium Pro (P6) running at 150MHz. But it looks as if a four-way 620-based AS/400 (the 53S-2156) will have about 15% less power than a four-way P6-150 server. Crafty tuning and cranking up the clock speed on the AS/400 could put the AS/400 on par with P6-150 systems, but Compaq Computer Corp and other vendors, including IBM, will likely get P6 servers running at 180MHz and 200MHz to market long before IBM is ready to announce a refreshed line of AS/400 RISC systems. I

BM has obviously had more problems with the PowerPC 620 and its progeny in the AS/400 line than it has had with the PowerPC 601 and 604 chips used in RS/6000s. But we think that there is another culprit behind the RISC AS systems coming up short: bloated system software. IBM has not made radical revisions to AIX for years. OS/400, on the other hand, is going through its first major rewrite for the RISC AS engines. (Think of v3.1 as a dry run for v3.6). As IBM gets v3.6 into shape, it will tune it so customers can squeeze more power out of their RISC AS/400s than they can right now. But this will take years. Today, it is more cost-effective to buy a Pentium server to support work that doesn’t have to be built under OS/400. The price-performance curve in the mid-range has a much steeper slope than IBM and many other vendors thought it would have. Upstart vendors like Compaq are supplying cheap, powerful and well regarded mid-range machines that rival products from IBM, Digital Equipment Corp and others. Hewlett-Packard Co and Sun Microsystems Inc are right there alongside Compaq, too. If the mark was $500 per TPC-C TPM when the RISC AS/400s were announced, it will be under $250 by the time they actually start shipping to customers. Compaq already offers Windows NT systems that dip below this price point. The situation will only get more difficult when Compaq starts shipping P6 servers this year. IBM will have a tough time selling even Advanced Server models to existing customers for new applications because they are considerably more expensive than P5 and P6 servers running NT. And there’s only so much that IBM can do to bring the cost of AS/400 hardware down. The myriad personal computer server vendors sell more than 2m units a year. They use many of the same high- volume, low-cost components that are put into personal computers, of which some 60m a year are sold. IBM is lucky to sell 60,000 AS/400s a year, and they use unique components that almost certainly have much higher unit costs.

Just as proprietary

Similarly, the uniqueness of AS/400 software, which has played into IBM’s favour in the corporate market thus far, will weigh against it as Windows NT becomes a popular system for mid-range applications. Windows NT is perceived as being an open system, even though it is just as proprietary as OS/400. And while there may be an impressive 300,000 AS/400 licences out there, Microsoft sells that many new NT licences in a single year. AS/400 managers, well aware of the price gap between their primary systems and fast personal computer servers, will have an increasingly harder time justifying their AS/400 budgets. From now on, managers will have to be smart about what work is appropriate for an AS/400 given the current economics of the mid- range and what is not. The potential savings from using NT on Pentium boxes in conjunction with AS/400s is going to make it well worth a serious investigation. Copyright (C) 1995 Technology News of America Inc. All rights reserved