XSM and Clustering: The Future of OS/400 High Availability

Published: May 23, 2006

by Alex Woodie

With improvements to resiliency features such as Cross Site Mirroring (XSM) in the latest release of IBM's OS/400, the scope of high availability options available to iSeries users is increasing quite rapidly. While many iSeries users aren't fully aware of what XSM is or how it can benefit their businesses, IBM is clearly steering users and its high availability software vendors down the road toward XSM-based replication and clustering as the future of high availability.

XSM is a high availability technology that IBM introduced with OS/400 V5R3. The technology, which is a form of clustering, provides another way to move application data and objects from one server to another. On a technical level, XSM works by mirroring the contents of one independent Auxiliary Storage Pool (iASP) to a second iASP running on a remote machine. One of the advantages of XSM over switched-disk (also based on iASPs) is that the connected iSeries do not have to be on the same HSL, which breaks geographical boundaries. Also, since XSM is built into the operating system and the machine code, and doesn't rely on software vendors' technology, it should be more robust.

Mike Warkentin, the IBM alliance manager with DataMirror, says XSM is a promising technology for high availability. "IBM feels it's a little less complex to administer and maintain," he says. "It's almost a hardware-replication solution--it just works. You're getting the OS, everything."

The best candidates for implementing XSM technology are organizations with lots of Internet bandwidth, or those running very fast local networks, Warkentin says. "In a campus environment it makes sense, or if you have fairly large pipes," he says. "It's a synchronous process, so it needs a big pipe, or transactions can be delayed." Prospective XSM users should also have experience configuring iASPs, which "can be somewhat challenging to set up," he says.

IBM provides some facilities for managing an XSM environment in iSeries Navigator. However, users will get the most use out of XSM if it's incorporated into an overall high availability framework such as those provided by the high availability software vendors. IBM would also like its high availability vendors to support this new technology. However, not all of them have done so yet. OS/400 high availability vendors currently supporting XSM include iTera, Lakeview Technology, and Vision Solutions, which is planning to announce support for XSM in the near future

XSM is in DataMirror's development plans, but is not currently supported in its iCluster product. "We haven't heard a lot of customer asking for XSM. A lot of customer aren't really aware of what XSM is right now," Warkentin says. "We are looking at XSM right now to see if there's customer demand. It's a moving target." iCluster already supports OS/400 clustering, and iASPs, which will make it easier for DataMirror to support XSM, Warkentin says.

It is probably just a matter of time before DataMirror adds support for XSM. However, it may not be in the company's interest to support the nascent technology until there is a better indication of it's worth in the field.

There are several drawbacks to using XSM. First, it cannot be used to replicate some types of objects, because iASPs do not yet support them. Also, data held in remote iASPs replicated using XSM can only be used for disaster recovery. If users want to make use of that data for something else--such as running queries--they are better off going with a logical journal replication method from one of the HA vendors at this point in time, Warkentin says.

When these shortcomings in XSM are rectified, the obvious benefits of "hardware-based" replication will likely prove too tempting for vendors like DataMirror and OS/400 shops to resist.

A similar thing occurred with clustering. At the turn of the millennium, IBM was fervently pushing OS/400 clustering, but the technology was just not ready for prime time. However, clustering has steadily matured into a more usable technology, and IBM never backed down from applying pressure on its business partners, including iTera, which heavily resisted supporting clustering for months because it was successfully selling a simpler remote journaling-based product instead. However, because support for clustering was a prerequisite for inclusion into the business partner program that enabled iTera's resellers to offer double-digit rebates on hardware included in high availability deals, iTera relented and eventually supported clustering.

Clustering technology--including XSM--is clearly the future for IBM, its high availability business partners, and, by connection, iSeries customers. At DataMirror, more than one-third of iCluster users are currently using the product in cluster mode (the rest are using the logical journaling mode.) "iCluster is built on clustering, but you don't have to use it," Warkentin says. "I think there are a lot of customer using clustering and don't know it."

While IBM developed good technology in OS/400 clustering, it failed to convince iSeries ISVs to cluster-enable their applications. In fact, a recent search of IBM's global solutions directory turned up not a single OS/400 application that was cluster-enabled. However, the search function was so convoluted to use that it's quite possible this reporter just missed it.

Warkentin doesn't dispute that few, if any, vendors have modified their applications to take full advantage of clustering. But the important thing to realize is applications don't have to be cluster-enabled to wok with OS/400 clustering and to gain the advantages of clustering, including better proficiency at handling communication failures. (Full cluster-enabled applications gain the additional advantages of faster failovers, and maintaining the status of applications following failovers so users don't have to log on again.)

Other vendors, including Vision Solutions, are having success at clustering, too. A solutions architect says the company is gaining proficiency in clustering three or more iSeries servers. "Three-node plus. That's what IBM is pushing," he says.

Steve Finnes, the IBMer who runs the OS/400 clustering and high availability programs, said in a recent Webinar sponsored by Lakeview that IBM and Lakeview have "collectively a large number of customers out there. We're very focused on clustering," he says. "It's really where the world is headed in a major way."

While XSM could someday spawn its own class of products, as remote journaling did, in the meantime, it's best to consider XSM as one more tool in IBM's resiliency kit.

"I think more shops will look at iASP and XSM as another method for providing disaster recovery and high availability," DataMirror's Warkentin says. "Not to the exclusion of high availability products. They will migrate down the road to adopt those technologies, just as they did with remote journaling and clustering."