The core value proposition of single-image clustering
Clustering has been used since the 1960s as a way to speed up parallelizable programs -- its use in mainstream commercial computing dates to the 1980s, when vendors such as Digital Equipment Corporation faced difficulties in scaling up their minicomputer hardware architectures to compete with mainframes.
At the same time, clustering turned out not to be so good at scaling linearly (e.g., add one processor's worth of workload and get only 70-80% of a processor's worth of added performance).
However, over the last 10 years, mainframe clustering has made significant strides in getting closer to linear scalability by better multiplexing of workloads between systems.
Moreover, the increasing popularity and maturity of virtual machine (VM) technology means each mainframe system can now support more than a thousand VMs.
Users have single-image clustering for up to 32 mainframes that effectively scale to what acts like one physical mainframe containing more than 30,000 VMs. That is 1,000 times the scalability of systems 10 years ago.
Today's clustering not only provides near continuous application availability via super fast failover communications lines, but it also offers a single-image, dynamic workload management and infrastructure simplification.
The cost advantages of single-image mainframe clustering are particularly important in today's recessionary times. It seems counterintuitive, but storage needs apparently continue to increase by 40% per year in the teeth of lower consumer demand. Meanwhile, database software is CPU-bound: To process the increased amount of information, users must scale the computing, not just the storage capacity. And that means that the need to extend mainframe scalability via clustering -- whether for handling server consolidation or Linux workloads, or just for increasing data-warehouse capacity -- continues to increase.
Parallel Sysplex and GDPS
Users can see the fruits of the integration between Parallel Sysplex and IBM's other business continuity technology, especially IBM storage/remote copy tools, in the latest iteration of Geographically Dispersed Parallel Sysplex.
The overall idea is to provide ultrafast communications between nodes in the cluster and between sites so that when something goes wrong in a very large operation, users can, in the worst case scenario, fail over in a couple of hours or less. More specifically, single-site operations promise zero downtime with no loss of transactions (not including site failures). Metropolitan two-site operations promise less than an hour of downtime with no loss of transactions; remote two-site operations promise less than an hour of downtime and loss of only seconds of transactions. Using HyperSwap to copy clustering's shared disks avoids disk storage being a single point of failure.
Other new features include a reduced-impact copy that diminishes copy overhead on operational systems dramatically (by 80% in some cases); coordinated Linux and z/OS recovery; health checks; and incremental resync. IBM has also added support for the Veritas open distributed clustering solution, but it is clear that the really high-performance case studies involve Parallel Sysplex (and HyperSwap), and that some organizations chose to use it for excellent performance in addition to business continuity.
Mainframe clustering and consolidating Windows apps
Recent announcements by Novell and Mantissa Software suggest it is possible to run many Windows applications, without code change, on the mainframe.
It may soon be possible to consolidate most of a large enterprise's apps on the mainframe.
That means that IBM server consolidation technologies would need a cost-effective, architecturally simple leap in scalability -- and that's just what Parallel Sysplex's single-image and clustered load-balancing can provide.
So, one action item for large enterprise IT seeking cost reductions in the face of compute-power demands is to look into consolidating Windows on the mainframe, and handling the increased demand via Parallel Sysplex.
Another interesting potential use case of Parallel Sysplex is where more than one mainframe is necessary but administrative costs are a concern. The point of a single system image is that the fewer systems you have to administer, the cheaper their administration is. While the total cost of ownership effects of clustering have not been studied as carefully as those of VM technology, it appears reasonable that clustering can reduce the administrative costs of distributed mainframes by quite a bit.
It remains to be seen whether that additional administrative cost reduction on top of VM cost reductions is important to the enterprise, but it is definitely worth investigating.
ABOUT THE AUTHOR: Wayne Kernochan is president of Infostructure Associates, an affiliate of Valley View Ventures. Infostructure Associates aims to provide thought leadership and sound advice to vendors and users of information technology. This document is the result of Infostructure Associates-sponsored research. Infostructure Associates believes that its findings are objective and represent the best analysis available at the time of publication.
What did you think of this feature? Write to SearchDataCenter.com's Matt Stansberry about your data center concerns at firstname.lastname@example.org.
This was first published in August 2009