In June 2009, an IBM executive delivered a System z presentation that provided a simple prescription for fit-for-purpose decisions. The new hybrid zEnterprise 196, a mainframe that can incorporate blades running x86, POWER7, AIX, and Linux along with the usual System z assist processors (zAAP, zIIP, and IFL) under unified virtual platform management (the Unified Resource Manager) complicates the fit-for-purpose calculation considerably.
Fit-for-purpose refers to the decision process one goes through in determining which workloads should run on which platforms. In June 2009, the choices as presented were pretty straightforward. You chose mainframes and high end UNIX servers for high volume transaction processing and big databases. UNIX platforms also were the recommended choice for business critical production applications, analytics, and high performance computing. Windows servers were left to handle web applications, web infrastructure, and collaboration. Based on the same thinking, Windows servers would take on workgroup productivity applications.
In IBM’s new fit-for-purpose scheme the four basic platform choice categories essentially remain but the application workloads are labeled Type 1, 2, 3, 4.
Type 1—mixed workloads updating shared data or queues
Type 2—highly threaded applications
Type 3—parallel data structures with analytics
Type 4—small discrete applications
As a hybrid machine—one able to cross the different IBM platforms—the zEnterprise 196 complicates the decision of where to put workloads because some workloads can be run on different platforms and in different ways within the zEnterprise. For example, you can run Linux using an IFL assist processor with or without z/VM or you can run Linux on a POWER7 blade or x86 blade within the zEnterprise extension cabinet, the zBX.
Similarly, you can run Cognos on the z on Linux using any of those platforms. The same goes for WebSphere. In the usual fit-for-purpose decision you consider data volumes and data proximity, the number of users, the number of transaction, usage patterns, and service and security levels.
But now you especially need to look at price and performance. At this point, IBM has been coy about talking about the price of the various zEnterprise components. They also have been silent about the real-world performance of the different components. In terms of performance are organizations to assume that Cognos running on Linux on z with data stored in DB2 on the same z will deliver the best performance?
What happens when you consider price/performance rather than performance alone? Will Cognos running on Linux on an x86 blade in the zBX extension deliver slightly less yet still good enough performance but at a considerably lower cost? Who knows? Until IBM provides more information about both pricing and relative platform performance there is no way to evaluate the tradeoffs involved in each fit-for purpose choice. Same with WebSphere or any other workload. (Further complicating the Cognos decision is the Smart Analytics Optimizer, a special blade for the zBX.)
Since some of the pieces aren’t actually shipping yet, IBM may not yet have settled on specific pricing. As for real world performance, we will have to wait for companies to begin deploying the new zEnterprise 196 in production situations that generate meaningful production and performance metrics. Only then can organizations make good fit-for-purpose decision with the new machine. As has been noted here before, the zEnterprise is a work in progress. Stay tuned.