% fortune -ae paul murphy

PA-RISC and the freedom to choose

Back in the early to mid nineties HP's K series machines offered real value to customers: combining mid range performance with extraordinary reliability and good support to earn deep customer loyalty. Since then, of course, HP was first taken over by Wintel true believers who shut down the PA-RISC line the company's enterprise computing reputation was built on, and is now in the hands of cost cutters. As a result HP-UX is now out contention for new installs and die-hard users face real problems.

In most cases the main problem is simple: there's a lot of old HP gear out there that still works, and the more painful it was to get things working smoothly, the more resistance there is to undertaking new adventures using a different architecture.

Unfortunately push has come to shove with HP no longer making the PA-RISC and the cost of used parts for upgrades or replacement just now starting to escalate. Thus an HP 8600 CPU set that sold for $180 two years ago is now worth about $240 and board assemblies that used to be throwaways are drawing real bids on ebay.

Talk to some of the people involved and what you hear is resistance and frustration: resistance to Itanium, resistance to Wintel, and frustration both with HP's failure to develop HP-UX and its own PA-RISC, and frustration with its refusal to reverse Compaq's decision to let Alpha and Tru64 wither and die. Talk to them about their alternatives and what I hear most is that they don't trust Sun and miss the good old days when IBM defined both computing and the vendor-customer relationship.

There are historical roots for both of those responses. On one side, many of the organizations involved, and the older people now in charge within them, came to HP as emigrants from the IBM world - and have the same wistful longing for the good old days in the country of their youth that we see in older immigrants anywhere. And, on the other side, Sun was the bad boy competitor they didn't pick when they went to either DEC or HP, and the fact that history has proven them wrong hasn't endeared the company to them.

So what to do? Well, HP's support for Linux on Intel as the alternative to HP-UX on PA-RISC or Tru64 on Alpha has had the odd side effect of making migration to IBM, Sun, or just plain white box Lintel much easier.

Specifically, what's happened is that HP's dwindling market share in big servers has led its software development community to hedge by moving their products to either, or both, Linux on Intel and/or Solaris on SPARC. As a result a majority of the commercial software once available for HP-UX or Tru64 is now available in later releases on Linux, and, of course, anything that runs on Lintel is trivially portable to Solaris on x86 too.

Equally importantly, what support HP has provided for open source porting to HP-UX and Tru64 has produced a list of about 480 open source packages for HP-UX -virtually all of which are also available free of charge on both Linux and Solaris.

In other words, commercial or open source software dependencies are no longer much of an issue with respect to the timing of a decision to leave PA-RISC (or Alpha) for either Linux or Solaris. That leaves cost and custom developed software as determining issues on the rational side of that decision.

As little as six months ago, moving from a higher end HP box to something of comparable power from Sun or IBM meant spending significant dollars: from a few hundred thousand dollars to replace the 8 to 16 way units to several million to go from a 64CPU HP complex to a 12, 15, or 25K Sun box.

However, Sun's new generation Opteron and UltraSPARC T1 servers are particularly good at many of the same things HP-UX was good at on PA-RISC: offering high reliability and quick processing on smaller jobs that don't need massive amounts of main memory. As a result the hardware side of the cost barrier facing most users in this position has dropped considerably: a Starfire 25K with 144 fully symmetric USIV cores accessing 512GB of RAM will still cost several million bucks, but a small Sun Opteron or T1 server can now blow away most of the older HP gear at the cost of three months in HP support - and comes with a one year on site warranty.

Notice too that disk costs have fallen almost faster than CPU costs. That terabyte Emulex or HP RAID that took so much effort to get to operational acceptability in 2000, can now be matched with no real software setup effort for under ten thousand using ZFS and US320 with the T1 or about twice that as a Fibre Channel array with a pair of smart gigabyte buffers using Linux.

What all that means is that it's now down to custom code as the only remaining real barrier to the change - and most of that should probably be re-evaluated against commercial alternatives anyway.

So here's the bottom line for PA-RISC (and Alpha) users: recent price/performance changes at Sun, coupled with long term disk cost reductions and widespread growth in open source, mean that any analysis you did more than a month or two ago needs to be revisited, because the rules have changed, and your escape opportunities have changed with them.

Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specialising in Unix and Unix-related management issues.