On August 29, 2012, IBM announced a new version of its venerable mainframe architecture — the zEC12. Along with this announcement IBM introduced two new products: Flash Express and zAware. In this report I describe what Flash Express is, what it does, and I show some early results of the new facility.
What Is Flash Express?
Flash Express is all about using solid state drives (SSDs – these are optional) to improve the performance of certain mainframe tasks.
The way I see it, a bunch of IBM developers got together one day and thought about which management workloads could best benefit by using fast SSD’s as opposed to slower, mechanical DASD (direct access storage devices). What I think they were looking for were situations where programs become slowed when having to read/write to mechanical disks. And these developers immediately came up with two scenarios where SSDs could significantly improve performance: 1) in paging; and, 2) in dump processing.
Paging is an activity where code is loaded and then immediately “paged-out” as a workload attempts to reach steady state. Paging-out to mechanical drives takes a lot more time than paging-out to SSDs because mechanical drives have moving parts and slower read/write speeds than solid state, no-moving-parts SSDs. By using speedy Flash (SSD) drives, it takes the paging subsystem a lot less time to reach steady state as compared with slower mechanical drives. By reducing the amount of time that an operating system has to spend in paging activities, enterprises are able to spend more time doing useful work. The time saved can reduce a paging exercise from a minute to a number of seconds – but in mainframe environments where many applications are being swapped out as a business resets from one activity to another, recovering a number of seconds can be a pretty big deal.
For an example of why cutting transition times down is so important, consider the following. Flash Express is particularly useful in helping with workload transitions – for instance, a system running an overnight batch workload may need to shift to online transaction processing at the open of business on a given day (banks and financial institutions such as stock markets frequently halt batch jobs in the morning to transition to transaction processing and call center support workloads). Further, in the future, Flash Express may be used to speed paging when handling new Big Data workloads. By speeding up these transitions, enterprises can get more useful work out of their mainframe servers.
Flash Express will also be used over time to speed up SVC (supervisor call) dump processing by reducing latency times. In short, when work needs to be analyzed it needs to be dumped and pulled in from disk such that a snapshot of an occurrence can be isolated. As was the case in the previous paging example, SSDs enable data to be processed by the dump more quickly. And, as a result, managers and administrators are able to accomplish management tasks more quickly, leading to faster troubleshooting and better performance. This, too, may seem to be a mundane use for SSD technology, but to systems administrators (such as a group of administrators in Italy who apparently got very excited by this), speeding up dumps can also be a very big deal. In this case, Flash Express is a big deal because, while the dump is processing, transactions can be stopped.
As for the future of Flash Express, I expect more innovative uses for SSDs in the mainframe to surface. Any workload that can benefit from significantly faster reads and writes is a prime candidate.
Translating Flash Express into Real World Savings
The primary value proposition of Flash Express is that it can improve performance and availability. Some of the early results from real-world test scenarios that have been run by IBM indicate that DB2 can perform significantly better using Flash Express. Upgrading to an EC12 gives DB2 a substantial performance boost for a number of reasons (an up to 25% performance boost); while tests are also showing that a performance improvement of up to 3% can be seen with Flash Express on the EC12. In other words, DB2 on the zEC12 gives users 25% more throughput as a base (due in large part to the faster CPU), plus an additional up to 3% performance boost from Flash Express, yielding up to a 28% improvement in DB2 performance. Better performance means more efficient computing – and more efficient computing means lower cost to compute.
For more information on Flash Express, readers should be aware that IBM will soon publish a new, in depth white paper on Flash Express. Additionally, a technical write-up on Flash Express can be found in IBM’s zEC12 Redbook (located here — see Appendix C). Further, the SHARE user group will feature a technical discussion on Flash Express at its user group meeting scheduled in San Francisco, California (a link to this précis of this discussion can be found here).