September 2004

HP updates storage issues in Chicago

MPE/iX to handle bigger disks for 3000s, while HAFO improves availability

By Steve Hammond

Although its end-of-life date is not far over the horizon, HP continues to offer availability and disk expansion opportunities for HP 3000 users.

In a session at HP World 2004 in Chicago, HP’s Jim Hawkins detailed some of the improvements they are making to MPE regarding storage.

In response to the Number 8 item on the 2003 System Improvement Ballot, HP has investigated the issue of the 300GB limit on size of hard drives. “We have done some testing,” Hawkins said, “and we were able to determine that MPE can handle a drive up to 512GB.”

The current limit remains 300GB, but in the near future, HP will be issuing patches to allow the use of larger disks. Unfortunately, 512GB will be the maximum disk size, because increasing that limit would require significant operating system changes. So ultimately, even when you connect a drive larger than 512GB, only 512GB of space on that drive will be utilized.

Hawkins also addressed the question of device driver hardening. “We are currently investigating what needs to be done to harden the MPE drivers,” he said. “It may not be easy, but when we stop supporting MPE, we want to be able to say what needs to be done to harden the drivers, if it can be done.”

Another issue still under investigation is a SCSI pass-through driver for MPE. No decision has been made on this one, only that they will continue to look at “whether such a driver will be easily do-able.”

Hawkins said in the session that even though MPE is nearing its end of life, HP still has people on the team trying to make improvements for users.

Higher availability

If you are thinking about clustering and high availability for your 3000, then Walt McCullough has your back. Walt has been with HP for 23 years, almost all of it in the MPE realm. He now is the MPE/iX High Availability R&D Architect, and he gave details at HP World on how you can keep your 3000 applications functioning in unplanned downtimes.

The two solutions offered by HP are Cluster/iX and HAFO/iX (High Availability Fail Over). Both have their expenses, hardware-wise, but this is an expense that must be balanced against the possible business losses when unplanned downtime occurs. A study done in 1996 estimated downtime costs for a brokerage house will run between $6 and $7 million an hour, while a large retail enterprise can lose almost $3 million an hour in credit card sales. Even a small catalog sales company can drop $100,000 in that hour. Therefore, setting up a high availability environment can pay for itself in a single outage.

McCullough did note in his presentation that running both these products in the same environment is an unsupported configuration.

Cluster/iX creates an environment that protects against controller failure, severed cabling, FC switch failure, HBA failure, software/application failure and OS failure. In combination with Continuous Access XP, it will also help you keep your data and applications available in instances of site power outage, catastrophic equipment failure and will create a disaster site.

It is best implemented when the user community can tolerate minimal downtime — not enough time for a system reboot (one to 10 minutes), in an environment where you have some control over the applications and are willing to make some changes to accommodate the clustering and where the users only want one copy of the data. It requires two servers, implementation of some scripting (which HP can do for an added cost) and a higher level of administrative training. (In other words, if you do things wrong, data corruption will occur.)

In short, there are two 3000s, with a single set of disks or arrays that are shareable and recognized by both servers. Scripts are running on both servers — a script on the “primary” creates a “heartbeat,” which a script on the “secondary” listens for. If that heartbeat fails, then the secondary takes over — and the script can also be written to page someone and also continue to monitor the primary to determine when its heartbeat returns. Switching control back is also part of the scripting.

HAFO/iX addresses the same problem from a different direction — you have one server, but redundant disks. It protects against controller failure, severed cables, HBA failure and FC switch failure. It is set up with dual active paths to both “sets” of disks. HAFO/iX detects a component failure and redirects the data to a redundant path.

Like Cluster/iX, HAFO/iX comes with a set of caveats — HAFO/iX will not work in conjunction with Cluster/iX, all logical devices must use similar connection technology, and it is not fault tolerant — meaning any unplanned outages may become planned outages to fully recover. HAFO does add significant complexity to the operating environment (which requires a far better understanding of the system characteristics and better planning for any system changes) and it is highly dependent on performance expectations, which means a high potential for false failovers.

McCullough concluded with his own caveat — HAFO/iX is not some magic panacea of high availability. It should not be implemented on a system that has inexperienced or part-time system management. This is a complex environment that needs monitoring and tuning — false failovers will occur if the system is not closely observed. It solves problems, but the solution comes at a price management must be willing to absorb.