Hardware forecast, migration plans sketch system to 2002 and beyond

Platform planning manager Dave Snow discussed a handful of slides marked “Subject to Change Without Notice,” but the caveat was applied to some products which are already shipping as well as 21st Century releases. The Commercial Systems Division continued its tradition of briefing customers on far-forward platform plans in a talk that last year identified IA-64 as a candidate technology for HP 3000s.

The earliest technology refresh in hardware will come from a 240 MHz PA-8200 for the midrange 9x9 systems by the end of 1999. HP will also support 12-way 997 systems in the same period at the top of the performance lineup.

HP will be able to take its 997 systems to 12-way processor capabilities in 1999, Snow said. But more significant changes to MPE/iX must take place before 16-way to 32-way — or even 64-way — HP 3000s hit the streets. HP also has scaling work to do on getting to 32Gb memory support for HP 3000s. Snow said HP is investigating scaling issues such as “tables being scanned linearly through the operating system, and the tables are getting bigger as you add more memory,” Snow said.

PA-8500 plans

Snow had outlines for a generation of HP 3000s that will use the next iteration of PA-RISC processors, the PA-8500. HP will press customers into a box swap to use the 8500s, and this generation of 3000s will drop support of HP-IB altogether. HP has already eliminated HP-IB support on the 9x9 line of 3000s. HP-IB peripheral support “was great while it lasted, but it’s not going to be here much longer,” Snow said.

The first major change in HP 3000 hardware will take place in preparation for using the IA-64 technology. Customers are facing a box change to get to the 8500 processors — sometime in Year 2000 — because the current processor bus backplane performance won’t scale to match chip horsepower. “We have to increase the performance of our buses,” Snow said. The next generation of 3000s will do PA-RISC and IA-64 in parallel, so buying one system will serve for both kinds of processor boards.

Higher numbers of processors and greater memory support were among seven key platform technologies Snow identified as being developed or investigated for the 3000. HP also included the high-performance crossbar parallel backplanes on its list, as well as PCI and FibreChannel IO busses; networking links to include ATM, FibreChannel and 1-gigabit LANs; FibreChannel and Ultra-SCSI support for device connections; and I2O I/O software architecture for downloadable drivers.

Advanced device support

Despite all the talk about FibreChannel, 3000 customers will be waiting on its arrival for a few years. As a result of the growth work, the division is working on next generation platforms’ IO instead of supporting High Speed System Connect busses (HSC) IO interfaces for 99x and 9x9 systems. This will delay FibreChannel support for 99x and 9x9 current systems beyond HP’s 1999 projection, announced last year.

Snow explained that “our first introduction of FibreChannel will be on the next generation platforms. We have decided to work on next generation platforms before we complete doing anything in the FibreChannel/HSC world. We are still looking at whether it makes business sense — in the timeframe of 2000 — to also bring the FibreChannel bus back to the current platforms. We’ve not made a commitment to do that at this point.”

Snow said the 3000 really “needs higher buses than HSC. The industry is moving toward PCI; not just PCI you might get on a PC, but times-two and times-four PCI.” These high-speed interface cards will require a high-speed interface to the devices themselves, a place where Ultra-SCSI is being investigated for HP 3000 use.

Near-term technology for networking is ATM and FibreChannel, but Snow said “very quickly we see on the horizon gigabit Ethernet LANs coming down the pipe. That’s probably where we’re going to focus our first effort — allowing you to reuse the cable you’ve already put in for 100 megabit LANs, in the 2000 timeframe.”

I2O “is a whole software architecture with tremendous potential, which allows anyone to write a driver that’s put on a PCI IO card, plug it into the system and have it downloadable into the system.” The standardized architecture for IO cards executes drivers off IO cards. The 3000 division is investigating what’s needed to put a stub driver into MPE that would pass messages to intelligent IO card drivers.

The potential benefit of I2O gets third parties to supply both IO cards and drivers, so HP doesn’t have to delay IO support as it writes drivers for new IO devices. Off-the-shelf IO cards from places like Fry’s Electronics could be used in HP 3000s if I2O earns a spot in the system’s technology lineup. HP “would probably end up doing some certification on our side, but it has real potential in reducing IO development costs and time-to-market for IO cards.”

New buses on track

Three new platforms will emerge in the 2000-2001 timeframe using some or all of these technologies. HP will provide a 30- to 40-percent increase over the 12-way Series 997s with the top range of these new platforms, using the PA-8500. HP will offer board upgrades to the 8550 and 8700 on these new platforms.

The new platforms will use intelligent buses for higher speeds, leveraged from the HP 9000 V-Class systems. “It presumes you have intelligent chips at any point that connects to a bus, which multiplex transactions across several parallel buses,” Snow said of the technology first leveraged from Convex. “Each bus runs at a gigabyte, and it looks like you have very high speed internal buses without all of the heavy duty investment into microwave technology for those buses.”

Fresh processor, new boxes

The midrange of the new platforms will use lower frequency versions of the 8500 with smaller IO and memory configurations. It’s in this timeframe — beyond the Year 2000 — when the low-end of the 3000 line will get its first processor refresh. The low end will be using the PA-8500s with still smaller IO configurations.

High end HP 3000s get the first 8500s by the end of 2000, with midrange systems to follow and low-end 3000s gaining the processor after that. “We’re going to roll the entire product line in 2000 and 2001,” Snow said.

Software IA-64 support

Even though these new systems will have the hardware capability to support IA-64, “we won’t have the operating system in place in 2000 or 2001 to support IA-64,” Snow said.

He reiterated HP’s support for binary compatibility in IA-64, so “any Native Mode applications will move forward into the IA-64 world.” Possible exceptions for support may be if today’s application calls a data structure in the operating system. “That might have some problems, but in general if you play by the rules of the game, the PA-RISC stuff ought to move forward seamlessly,” he said.

HP is still deciding what to do with software written for the Classic MPE V systems in IA-64 support. “That’s emulation on top of emulation. It’s not outside the realm of possibility, but I’m not prepared today to commit to doing that,” Snow said.

“It would be a nice marketing message to say that code written 25 years ago could run on IA-64, but we have to evaluate the cost of doing that against the benefit of that message. That’s four years downstream, so you have an opportunity to continue to move away from that code.”

IMAGE/SQL will run natively on IA-64, and there will be compilers producing IA-64 code to coincide with the first release of the MPE which supports IA-64. “Which of those compilers [will make the transition] is still an open issue at this point,” Snow said. “In an ideal world we’d like to have a complete set, but it’s not clear we will have a complete set in our first release. Part of this has to do with how quickly we get the first release out. It’s not clear that a pure Pascal compiler is an absolute must-have at first release, for example. I’d guess COBOL probably is a must-have.”

Snow concluded by flashing the slide projecting 3000 performance that was first shown this spring in the IPROF Programmer’s Forum. HP believes it will be able to deliver performance 10 times greater than today’s 3000s over the next 10 years, “including the PA-RISC stuff, and at some point there’s a crossover to IA-64.”