The long wait is over. On July 22, IBM announced the next generation of mainframe processors. IBM officially calls the new processor the zEnterprise 196, or z196, machine type 2817. The new machine boasts the usual increases in size and performance as well as something new.
ZEnterprise processor details
A zEnterprise 196 cabinet can hold four books, each book containing six quad-core processors and up to 786 GB of memory with four levels of cache. The biggest and baddest zEnterprise sports 96 cores, of which 80 are available to the customer.
As with all the previous generations, the z Enterprise 196 is faster. IBM says the microprocessors run at 5.2 GHz as opposed to the z10's 4.4 GHz. This translates to a 40% capacity boost over a z10 with the same number of engines, while a fully loaded z196 contains 60% more capacity than a fully tricked-out z10. Your mileage, of course, will vary.
It sounds like some of the internal pipelining was also changed, as the data sheet claims there are new instruction execution sequences for better performance. For the application folks, IBM has compilers with optimization options for the new hardware, some of them presumably taking advantage of the 100 new z196 instructions.
A fully loaded z196 contains just over 3 TB of memory and allows for up to 1 TB per logical partition. This model also introduces something called Redundant Array of Independent Memory (RAIM), which appears
IBM made several changes to make the z196 more "green." For instance, it uses less power and allows I/O cables to go through the top of the case so as to allow more air to circulate below the floating floor. In addition, this process reintroduces water cooling, which is supposed to be more energy efficient than pushing cold air around. Finally, the new processor can use high-voltage DC power, which should reduce some electrical infrastructure requirements. The last two features are very nice, but a lot of data centers may not be set up to take advantage of either immediately.
Why zEnterprise is a big deal
But additional processing power doesn't appear to be the main thrust of the z196. IBM devotes most of the marketing material to telling us the new mainframe can contain outboard specialty processors and Power7 engines on zEnterprise Blade Extensions (zBX) running AIX or Linux. Included in the announcement was a statement of direction to include X (x86) processors as well. The blades are tied together and to the mainframe through two internal networks, one to manage the configuration and the other for data.
According to IBM, these are the advantages to sticking these blades inside the mainframe's cabinet:
- The blades can participate in the cabinet's power management
- The server hardware and its network are more physically secure
- Locating the servers inside the box speeds up the interaction between the different platforms and puts distributed applications physically closer to mainframe data
- Putting the servers in the same cabinet allows z196 to manage application workloads from one end to another
At first glance this sounds like IBM's entry into the "data center in a box" concept with the happy addition of the mainframe. But the z196 is a little more ambitious.
Putting the blades in the cabinet allows features like Smart Analytics Optimizer (SAO) for DB2 to work better. SAO allows mainframe DB2 to offload huge queries onto other engines. There, the queries can churn away in parallel before returning the answer to DB2. There are some broad hints in the announcement that there may be other types of these accelerators to follow, possibly on special processors outboard of the mainframe.
The last bullet point alludes to the new z Unified Resource Manager. Unified Resource Manager is supposed to able to see application performance from end to end and make adjustments if necessary, sort of like a global workload manager (WLM).
IBM's vision may consist of one big Z cabinet containing mainframe processors and blades supporting dozens, if not hundreds, of virtual servers controlled by a hypervisor. In IBM's case, the nominee would be z/VM. Inside of each z/VM instance is a WLM-like agent that reports on server performance to Unified Resource Manager and relays resource allocation resource recommendations to z/VM.
Z Unified Resource Manager, running somewhere, would take the performance information from its agents and apply the knowledge of application topography and goals. In turn, it would make suggestions to the hypervisors about which virtual servers may need more resources and those that could do with less. Basically what we have here is WLM writ large across disparate platforms.
I'm not sure if IBM is there yet or even if cross-platform performance management is its ultimate goal. However, it's something it could evolve into if customers buy into it. If it does come to fruition, this setup is kind of a demotion for z/OS, as it is just another virtual server inside the box.
ABOUT THE AUTHOR: For 24 years, Robert Crawford has worked off and on as a CICS systems programmer. He is experienced in debugging and tuning applications and has written in COBOL, Assembler and C++ using VSAM, DLI and DB2.
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 2010