Last week we reported that late-generation XP storage arrays from HP work with the HP 3000. Two system integrators supplied more details on how to make a beast like the $14,000 XP20000 serve an MPE/iX server -- along with other hosts running more popular operating systems. HP ran a YouTube video back when the system's top-end was the XP12000. The video was called Bulletproof, featuring an array that continued to work after it was shot with a high-calibre rifle.
Craig Lalley pointed out some warnings about these new XPs. He called them catches, as in, "there's a catch."
The last XP array sold when the HP 3000 was given HP's death sentence was the XP1024. In order for the 3000 to talk to and control the XP array — i.e. split the mirrors, resync the mirrors and mirror status — the HP server uses a piece of software called RaidMgr. It is on every HP 3000, and it goes out with Posix. You find it in the account /tmp/raidmgr
However, the newer systems, XP10000 and up, require a different version of RaidMgr. Usually it can be found on the XP array CD. That CD holds all the firmware and documentation.
Pivital Solutions also sells and supports disk arrays. The company aids 3000 sites through MPE/iX software contracts as well as hardware service. Steve Suraci of Pivital had stronger reservations about using the XP20000 and XP24000.
"In theory, I can’t see why not — but in practice, I would be a bit hesitant," he said. "I’m not sure that a customer big enough or savvy enough to want one of these would be a good candidate to be a guinea pig, testing a theoretical solution."
Suraci said the reach for this newer hardware to support 3000s reminded him "of all the crashes we used to see on the 9x9s using the third-party fiber/SCSI bridges. In theory, they were touted to be a reasonably-priced solution that should have been foolproof. In some cases they were, in others the customer paid dearly in downtime for an unfixable problem."
Being a support company, Pivital's main mission is to keep its customers from unplanned downtime or surprises with storage.
Lalley's view is that "all the XP arrays support MPE. The OS talks to the array via a LUN that is configured in MPE as a Control device. Raidmgr is incredibly fast, and so much faster than working directly on the array."
The benefits for maximizing storage flexibility are out there—potentially.
Raidmgr is not required for the HP 3000 to run on the XP array. It's only required to allows the HP 3000 to control the array. If the HP 3000 is only using the disk with no special software, then Raidmgr is certainly not required. But it is a big bonus for Business Copy and Continuous Access.
A lot of the XP arrays have multiple hosts, not just the HP3000. The XP arrays are enterprise class system. Literally thousands of hosts as virtual machines can be configured on one XP array.
Suraci was cautious — in part because of the difficulty of testing the device with a 3000 to assure a customer these big arrays are production-worthy.
I think anyone that really needs this solution to work better be ready to pay the consequences if it doesn’t. It’s real hard to road-test production and replicate all the real-life scenarios that could lead to the solution's demise. Just because the 3000 sees it and you can configure it, does not mean you want to trust your production environment on it.
It's a pretty good bet that the production budget for an MPE/iX server in 2016 doesn't have a $14,000 disk array chassis on the ledger. But if the biggest XPs could be connected to a 3000 along with other servers, then perhaps MPE/iX support for the arrays would be a bonus.
It would not be the first time an official HP status on 3000 disks turned out to be something that field testing disproved -- a raft of third-party SCSI disks were suitable for MPE/iX, back in the day with SCSI was a current standard. That testing for the indie SCSI disks took place on lab benches, though. Adoption in production environments came later.