Nov 13 ’12
The zEC12: Fast, Robust, and Lower Cost
When will people stop marveling that the mainframe isn’t actually dead? At the end of August, IBM introduced its fastest mainframe ever. The new machine, the zEnterprise EC12 (zEC12), delivers more performance and capacity than its predecessor but at a lower price/performance. It also adds two major new capabilities, IBM zAware and Flash Express, and supports hybrid computing.
The z196 was a breakthrough machine that introduced multi-platform hybrid computing through the Unified Resource Manager (zManager) and the zEnterprise BladeCenter Extension (zBX). The zEC12 is more like the next revision of the z196 than something strikingly different. With a 5.5 GHz core processor, up from 5.2 GHz in the z196, and an increase in the number of cores per chip (from four to six), the zEC12 is faster. Supporting 101 configurable engines compared to 80 on the z196, IBM calculates it delivers 50 percent more total capacity in the same footprint.
The two new capabilities, IBM zAware and Flash Express, promise to be useful, but neither is a game changer unless you’re finding it difficult to keep your z machines running or are experiencing paging problems. If you didn’t already have a zEnterprise, these capabilities alone likely won’t spur you to acquire one.
IBM historically has released the next mainframe three years after the previous machine. The zEnterprise was introduced in July 2010. Based on that schedule, the next machine would be expected in 2013. Industry buzz for the past year was that the machine was coming sooner, maybe late in 2012.
So why now? Consider the latest IBM financials. In the second quarter of 2012, revenues from System z mainframe server products decreased 11 percent compared with the year-ago period. Total delivery of System z computing power, as measured in MIPS, decreased 8 percent. Both Power and System x revenues were down also, but not as much.
IBM has continued to sell new zEnterprise machines, just not fast enough. Organizations typically wait to upgrade until their next normal purchase cycle, often based on a three-year lease unless they’re offered an obscenely good deal to move sooner. By bringing out the new machine now, combined with aggressive pricing and promotion (such as deferred payments until 2013), IBM can jumpstart new sales a full quarter or more early.
To be fair, the zEnterprise overall sales performance hasn’t been that bad. IBM reports that, as of second quarter 2012, it signed up more than 140 new accounts since the z196 launch and sold more than 150 zBX units. zBX adoption, a key indicator of the acceptance of IBM’s hybrid computing strategy, is happening faster than the initial adoption of Linux on System z in its early years.
There are some mainframe shops that suck up every MIPS they can get. IBM can count on these core customers to upgrade to the zEC12 right away. Plus, IBM is offering attractive financial terms for buying early and unusually aggressive price/performance pricing.
IBM zAware is a new analytics capability embedded in firmware with a built-in, self-learning feature. It’s intended to monitor the entire zEnterprise system to identify problems before they impact operations. It combines nearly constant monitoring of z/OS messages with analytics. With IBM zAware, you can avoid surprises and speed troubleshooting. It also will link up with Tivoli Management in the future.
Flash Express consists of a pair of memory cards installed in the zEC12; it amounts to a new tier of memory. Flash Express is intended to streamline memory paging when transitioning between workloads. It will moderate workload spikes and eliminate the need to page to disk, which should boost performance. If your shop encounters frequent thrashing when transitioning between workloads or can’t tolerate paging spikes, this will definitely help. Otherwise, it probably is of marginal value.
Speeds and Feeds
This is a stunningly powerful machine, especially coming just 25 months after the introduction of the z196. More and faster cores deliver a performance boost for all workloads and the system supports up to 3TB of Redundant Array of Independent Memory (RAIM), which protects against memory loss similar to the z196.
The processor has been optimized for better software performance, particularly for Java, Programming Language 1 (PL/1), compilers, and DB2. As with the z196, it handles out-of-order instruction processing and multi-level branch prediction for complex workloads. Large caches, almost two times more on the chip, speed data to the processor. In addition, Flash Express provides 1.6TB of usable capacity (packaged in pairs for redundancy, 3.2TB total).
New hardware functions also boost software performance. Specifically, the transactional execution facility, brought over from IBM supercomputing, delivers improved parallelism and scalability. A run-time instrumentation facility should reduce Java overhead. Two-gigabyte page frames, aided by Flash Express, can provide performance improvements for DB2 buffer pools and Java heaps. In addition, IBM boasts of a 30 percent improvement in IMS throughput due to faster CPU and cache, compilers, and other optimizations. Finally, a new IBM Enterprise PL/I compiler is designed to exploit the decimal format conversions facility for another performance boost. In addition, the L2 cache has been boosted 33 percent and the L3 cache has doubled, along with an increase in L4 cache.
The upshot: IBM estimates up to a 45 percent improvement in Java workloads, up to a 27 percent improvement in CPU-intensive integer and floating point C/C++ applications, up to 30 percent improvement in throughput for DB2 for z/OS operational analytics, and more than 30 percent improvement in throughput for SAP workloads. IBM has, in effect, optimized the zEC12 stack from top to bottom, starting with semiconductors all the way through compilers, the Java virtual machine, and a slew of optimized middleware. Database and Online Transaction Processing (OLTP) applications are certain to benefit, along with WebSphere and SAP.
Pricing and Costs
IBM is aggressive in its zEC12 pricing. Specifically:
• Hardware—20 percent MIPS price/performance improvement for standard engines and specialty engines; Flash Express runs $125,000 per pair of cards (3.2TB).
• Software—update pricing will provide 2 to 7 percent Monthly License Charge (MLC) price/performance for flat-capacity upgrades from z196, and Integrated Facility for Linux (IFL) will maintain a Processor Value Unit (PVU) rating of 120 for software but deliver more MIPS.
• Maintenance—no less than 2 percent price performance improvement for standard MIPS and 20 percent on IFL MIPS.
IBM will continue the System z Solution Edition Program with the zEC12, providing aggressive pricing for new workloads and deep discounts for first-in-the-enterprise customers. As always, understand the fine print; typically there are strict usage constraints.
In an analysis comparing the zEC12 with Oracle, IBM priced out three Real Application Clusters (RACs), four nodes per cluster, each node as a Linux on System z guest, and 27 IFLs at $5.7 million Total Cost of Acquisition (TCA) on the zEC12. The Oracle system required three Oracle RAC clusters, four server nodes per cluster, and 12 total competitive servers (Oracle hardware, no doubt) at $13.2 million TCA. The zEC12 clearly wins, but without the actual details, terms, specs and prices, who can really tell? (Note that TCA includes hardware and software acquisition costs for first year and then ongoing annual hardware maintenance and software support and subscription for the next two years.)
Assist Processors and zBX
Finally, the zEC12 comes with the standard array of assist processors, which are just configurable cores with the assist personality added. The zEC12 assist processors bring a MIPS boost.
The zEC12 also is a full player in IBM’s hybrid computing with the zBX a key piece. You will need a new zBX, the Model 003, to connect to the zEC12. Blade capacity and blade configuration between Mod 2 (z196, z114) and Model 003 hasn’t changed, so you can use the same blades. All that’s required is a code reload for the new zEC12.
As before, the zBX is managed through the zManager. In a new twist, System Director capabilities can be delivered through the zManager Application Program Interfaces (APIs). This will enable the ability to discover, inventory, and visualize zBX resources under the Systems Director umbrella and handle image management of zBX blades for provisioning of new virtual servers. It also will handle energy management of the zBX and power capping for System x and Power blades. Eventually, IBM intends to deliver workload-aware optimization for System x blades through the zManager and perform automated, multi-site recovery for zBX hardware components based upon Geographically Dispersed Parallel Sysplex (GDPS) technologies.
IBM has been telegraphing this machine for several months as the zNext. Although it delivers a substantial performance gain and price/performance advantage, along with new capabilities and optimizations, the most impressive thing about the zEC12 is the speed at which IBM delivered it, dramatically shortening its historic three-year release cycle. Should the industry expect a repeat in late summer 2014?