Silence is not always golden. In fact, sometimes it is more like nickel, which is just about all Hitachi Data Systems has been able to count on since IBM announced, re-announced and then ‘re- re-announced’ its G5 series of S/390 CMOS mainframes between May and September. Each time IBM talked about the G5s, they got faster and made Hitachi’s powerful and popular Skyline series of ECL mainframes seem unnecessary. Whether or not the G5s are faster than the Skylines is an academic question – close enough kills whatever edge Hitachi might have with its bipolar technology mainframes. The real question on everybody’s mind had been just what will Hitachi do? Will it deliver a new generation of Skyline processors, and will they, like their predecessors, take the IBM mainframe market by storm?
By Timothy Prickett Morgan
Yesterday, Hitachi Data Systems put these and other questions to rest as in announced its plans for its next generation of Skyline bipolar and Pilot CMOS mainframes. Hitachi says that the Skyline II ECL processors will be announced in the first quarter of 1999, with customer shipments expected in the third quarter. The company’s plans, as has been rumored for the better part of a year, call for Skyline II to deliver about just shy of twice the performance of Skyline; the Skyline is rated by IBM (the opinion that counts since much mainframe software is priced based on MIPS) at 140 MIPS, and Skyline II uniprocessor performance is expected at somewhere between 260 and 270 MIPS. Hitachi will also boost overall system capacity by a factor of 2.7 by going from eight-way (Skyline model 827) to twelve-way systems, giving an aggregate capacity of about 2100 MIPS, about twice the aggregate capacity of IBM’s top-of-the-line ten-way G5 Turbo 9672-YX6. Hitachi is betting that those customers – in banking and insurance particularly – who loved Skyline are going to go ga-ga over Skyline II. Overall system capacity is not nearly as important to these customers as is the raw speed of a single processor. Banks and insurance companies face huge nightly, weekly and monthly single-stream batch jobs, and the only way to speed up those batch jobs is to have a faster engine. Adding extra processors to a system won’t do anything unless customers parallelize their code by rewriting it to take advantage of system clustering and other parallel features such as IBM’s parallel sysplex. While many customers have done that, the amount of work and money to do this is enormous and the payoff small. Especially when customers can just throw Skyline (and now, G5 9672) hardware at the problem and let their programmers work on other problems, such as Y2K bug fixes and putting web and client/server extensions on mainframe applications. With Skyline II, high-end mainframe customers will see their batch windows shrink by almost a factor of two and this is extremely important; that magnitude of batch window shrinkage is what has driven the Skyline business for years. IBM is, once again, betting that Hitachi is wrong, that there aren’t enough customers who need a 265 MIPS engine to justify its development in expensive ECL technology. (IBM made this bet the last time around and Hitachi made billions of dollars proving IBM wrong.) IBM has to downplay Skyline II, seeing as its G6 processors, which the company hasn’t even begun talking about yet, will very likely not have anywhere near the raw uniprocessor performance of Skyline II. Hitachi may once again be right about what high-end customers want, and only time will tell. But the installed MIPS bell curve of the mainframe base tends to argue in IBM’s favor. There are simply a lot less customers who need a 270 MIPS engine than who were dying for a 140 MIPS engine. But there are still thousands of them nonetheless, and IBM and Hitachi and, to a lesser extent Amdahl, will fight over that money. Of course, Hitachi isn’t placing all of bets on Skyline II. Hitachi is, oddly enough, an OEM customer of IBM’s G series of CMOS mainframe chips, which it uses in its Pilot line of CMOS mainframes. Hitachi has announced five new Pilot P8 models, which it says provide better granularity than IBM’s 9672 G5 series. The new models include two new eleven-way CMOS machines, the B8T (G5 Turbo) and the B8S (normal G5), as well as a seven-way Turbo 78T and five-way (58R) and three-way (38R) reduced speed G5 models. Hitachi says that it will introduce its next generation of CMOS models in 1999, the first hint we’ve seen about IBM’s G6 processors, which are expected to use IBM’s CMOS 7S copper process and which should have about 35% more power than G5, coming in at about 200 MIPS, perhaps as high as 225 MIPS. Hitachi has reminded its customers and competitors that it is firmly committed to IBM’s parallel sysplex clustering technology even as it adds faster ECL machines; odds are, Skyline II is the end of the line for Skyline ECL technology. Like IBM, Hitachi is also trying to tap into user frustration over distributed networks. To that end, Hitachi announced a PC File Server program that runs in OS/390’s Unix shell and which provides print and file services that emulate those commonly run on distributed Unix and PC servers.