A CONVERSATION WITH NICK EARLE
HP ProFiles:
A Converstation with Nick EarleVP/Group Marketing Manager for HP's Enterprise Systems Group
In our September issue, we launched HP ProFiles with an exclusiveinterview of Lew Platt -- the big guy at HP. This month, Nick Earle, Vice President andgroup marketing manager for HP's Enterprise Systems Group, reflects on HP's computingdestiny. In San Diego at HP World, Nick and I talked about Merced, IA-64 and, ExplicitlyParallel Instruction Computing (EPIC) the once and future CPU architecture for the nextcentury.
I'm sure you'll agree that Nick was quite forthcoming about HP's advantages regardingIA-64: we'll have the best compilers. On Sun Microsystems: their users will be forced tomove to big endian Solaris. On IBM: we are making the HP 9000 look like MVS. And onCompaq: they finally have a good strategy, similar to HP's.
HP Pro: What's HP's value-add going to bewith Intel's IA-64?
N.E.: Everyone will have IA-64, becauseIntel's volume production means that you can't compete without it; particularly when youlook at the fact that it's going to run UNIX and NT faster than RISC and IA-32.
Three things that are going to be important in the future [with IA-64] are scalability,availability and manageability. And that's where HP will have a lead over our competitorswith IA-64.
HP Pro: What about performance?
N.E.: The performance of EPIC isdetermined by the parallelism you can inject into the source code at the static compilerstage. In the UNIX world, because you have a different binary for each version of UNIX,HP-UX object code is going to be tuned for an EPIC-based, parallel environment. So, we'regoing to have a performance edge [too].
HP Pro: What's Intel's position?
N.E.: Intel is desperate to point outthat it's a level playing field, but it's only a level playing field for SPECint andSPECfp [performance benchmarks], which is at chip level. Every vendor in the world usingIA-64 chips will have the same SPECint and SPECfp. But TPC-C, TPC-D or OLTP and datawarehousing are functions of system design and that's primarily driven by the compiler.
HP Pro: Why do you think you'll have thebest performance?
N.E.: We started with the compilers in1989 as part of a HP Labs project. By the time EPIC hits the streets, or with Merced --the first chip -- [HP] will be in the 11th year of compiler design for optimizing inparallel environments.
HP Pro: And scalability?
N.E.: Scalability, in many cases, isdetermined by the OS. When we launched HP-UX 11.0, we did it as 64-bit. But in fact, werewrote it to be engineered for EPIC. And the reason we could do that is because we'veknown what's in the specification since 1993-94. So, HP-UX 11 already has the EPIC-basedspecification embedded in it.
HP Pro: Can you give an example?
N.E.: Oracle is moving a lot of their[Oracle] code to Java-enabled components. When you are going to use Oracle apps or thedatabase you're going to go via Java Virtual Machines [JVM]. Because there are moreregisters to play with (in the EPIC architecture), that JVM stack could be up to 10 timessmaller. So, you can have a chip-based JVM. If you can get a 10 times smaller JVM, youhave the potential for 10 times the scalability.
HP Pro: But Sun isn't going with theirSPARC version of Solaris [to the IA-64]. Right?
N.E.: In Sun's case, they said that therewill be one version for IA-64 and it's the little endian version. And they are pretendingthat everything's OK.
HP Pro: But it's not OK?
N.E.: It's not OK, because Solaris onIA-64 is little endian, which means that the installed base [of Sun users] can't movetheir applications or their data. So, if you have 50 SPARC machines, you got all of theseprograms and say 3TB or 4TB of data all stored in big endian. Because the source codebetween big endian Solaris [for SPARC] and little endian Solaris [for IA-64] isincompatible, you've got to recompile every one of your programs or get a new binary fromOracle or SAP for everyone of your programs to move. That's not a trivial task.
HP Pro: And with UNIX to IA-64?
N.E.: Merced is bi-endian. So,RISC-to-IA-64 applications are going from big endian to big endian, so our [HP-UX] usersdon't have to change their data. Although Sun can take Solaris to IA-64, they can't taketheir installed base.
HP Pro: What if they port the SPARCversion, which is big endian?
N.E.: If you're Oracle, now you have twobinaries -- one for SPARC and one for IA-64. Or little and big versions. Now, look attheir [Oracle's] user base -- it's all big. Okay, [let's suppose] I'm going to drop mylittle endian software development activities and switch to big endian. Sun slimes SGI,NCR and Fujitsu who they've signed contracts with to take Solaris to IA-64. Which do youthink Scott McNealy is going to do?
HP Pro: Looks like a lose-lose situationall around.
N.E.: Personally, I think Scott McNealywill go off and do something else. At which point, everyone will blame Scott. In about 18months, you'll be reading that was Scott's view and now we can renegotiate. At the momentthat Merced starts getting traction Sun will take big endian over. But the disadvantage isthat they will have to start working on the compilers -- that is, embedding the IA-64logic into Solaris.
Sun is licensing Solaris as a product to get profit -- SunSoft, which has a P&L.HP-UX is part of the hardware division. So we're licensing HP-UX to NEC, Hitachi andStratus and what we get back is technology around scalability, high availability -- andnot money.
HP Pro: What's your take on Compaq'sstrategy?
N.E.: I believe Compaq's mission in lifeis to dominate the enterprise with NT. I don't buy the fact that they are gong to do itwith UNIX.
HP Pro: Why is that?
N.E.: They bought a struggling UNIXcompany called Digital Equipment with a 2.7 percent share of the UNIX marketplace. Theyare pumping money into Digital Bravo, the IA-64 version of Digital UNIX. And they havecross-licensed to Sequent. They will have a technically good UNIX on IA-64 to rival HP-UX.The issue for Compaq is that they haven't got ISV support.
HP Pro: So, the trick for Compaq is to get ISV availability?
N.E.: The trick for all of this is to getISV software availability. They are breathing life into Alpha, even to the point ofselling Alpha-based NT servers. But it's going to take more than that to bring the UNIXbusiness back to life. They are ripping out of Digital, the support services, consultingservices, the NT expertise -- which is very good -- and making a lot of nice noise to theVMS installed base. They finally got a good strategy, similar to ours.
But, they are going to have to grow it faster than HP, which now stands about eighttimes greater in market share. So, that means that they are suffering in terms of softwareavailability. Now why are ISVs going to produce the first version of their code forDigital Bravo rather than HP-UX?
HP Pro: What about NT?
N.E.: The NT business is very good.Digital was doing that very well. Compaq being Intel's biggest [customer] will clearly goto IA-64. Alpha is a hedge strategy until that appears. But the NT side [for Merced] isdifferent. Because there is only one binary -- then everyone has it.
Then your differentiation other than in your supply chain like Dell, comes down to thesame sort of things -- system design and first one to market. UNIX will be the bigdifferentiation for IA-64. There will be some on NT, but not as much.
HP Pro: What's HP doing for NT?
N.E.: We're trying to pull NT up [intothe enterprise], with R&D investment, better high availability and, where there is anopportunity, to bridge the gap between [NT] Wolfpack clustering and UNIX.
HP Pro: Final thoughts on the future ofNT?
N.E.: Fighting NT is like standing on thechair in the sand and shouting at the tide not to come in. It's tiring. You make a lot ofnoise and eventually you lose.
HP Pro: What's HP's strategy at the endof the day?
N.E.: When people talk about UNIX versusNT they got it wrong. NT isn't going to gobble up UNIX. It's really MVS versus UNIX --that's where we are going -- to take HP-UX and make it look like MVS. That's our goal.It's the second attack on the mainframe. The first was our surround strategy. But [themainframe market] is still $40 billion. It didn't die. First we put the wagons in acircle. Now, were picking them off with rifle shots.
We're making the HP 9000 environment look like MVS without the DB2 or CICS lock-in. Andwithout all the problems that come with the mainframe -- the proprietary technology, thebusloads of consultants from IBM, IBM's attitude.