Every two years IBM introduces a new System z mainframe that offers increased capacity and processing power — often followed by a scaled-down version for smaller enterprises and organizations. This year IBM introduced the System zEC12 — a model that offers 25% more computing power than its predecessor thanks to big improvements to the z processor, in cache, in the processor nest, and also due to new instructions. But, as usual, increased capacity is not the whole story. There is usually a lot more going on under the radar such as improvements in manageability, in resilience, in security — and even in system orientation. For instance, in years past, IBM has spent a lot of time and effort repositioning the mainframe as a highly-scalable Linux consolidation server.
The highlights of this year’s announcement include:
- Expanded Level 2 cache (by 33%) — and a doubling of cache in Levels 3 and 4 from 24 MB to 48 MB, and from 192MB to 384MB respectively. Cache serves data to the processor — so, with faster processors and more cache IBM is now able to serve more data to the processor, thus delivering more compute power.
- Instruction set improvements and new facilities enable certain workloads to execute more quickly (for instance, new Java workload performance can increase by 45%; DB2 by 30%; and PL/1 workload performance will also increase significantly as PL/1 exploits the new decimal format conversion facility).
- Other technology advances include new manageability, availability and security offerings — as well as advances in hybrid computing.
Probably the most important changes in this year’s System zEC12 are the amount of memory cache that has been placed near the processor. Notice the huge improvements in on-chip Level 2, 3 and 4 cache. What IBM focused on was putting as much memory next to the chip as it could logically justify — making this mainframe an impressive server for data-intensive computing tasks. Couple the on-chip cache with the ability to use up to 3TB of RAIM (redundant array of independent memory) and the mainframe becomes an impressive business analytics server…
This point about making the mainframe a highly scalable business analytics server is worth closer examination. With its increased capacity and processing power, the zEC12 is able to run data warehouse, analytics and OLTP applications using the same data store (instead of relying on multiple, snap-shotted, siloed individual databases). The ability to use the same data store simplifies data management (no need to worry about data duplication, for instance) — and it paves the way for enterprises to develop new insights. (IBM customers are now telling IBM that they are able to develop new insights that they could not have developed when using separate data silos). The ability to use a common data set across data warehouses, analytics, OLTP, and other operational environments is a distinct advantage for the highly-scalable System z.
Another very important change to the zEC12 can be found in the new transaction-execution facility. SHARE members should be aware that the current design of CMOS processors are reaching their architectural limit in terms of processor speed — and, at the sub-atomic level, processor speed is about to top-out. The mainframe System z processor is currently the faster microprocessor in the industry at 5.5 GHz — but CMOS technology is not expected to ever exceed 7 GHz. So processing speed improvements are going to have to come from somewhere other than the processor — such as in finding ways to improve multi-threading performance as well as through more optimal systems designs.
IBM’s new transaction-execution facility targets improving multi-threading performance. One System z user, SHARE member Ed Jaffe, describes this facility as having the potential to “be the greatest single advancement in multithread performance on System z ever. Every indication is that this new facility is performing beyond IBM's wildest expectations. The benchmarks I've seen so far have been extremely impressive.” As Mr. Jaffe “very soon, CMOS CPU 'cores' will get smaller, but not faster. If we can't go 'up' we must go 'out' and IBM is doing a great job of breaking down those barriers.”
Other important news regarding the new z is that the system’s specialty processors are faster and cheaper than ever before — and are more useful than ever before. At $66/MIPS for zIIPs and $36/MIPS for IFLs, these zEC12 specialty processors are just 17% of their original cost. SHARE member Cheryl Watson, president of Watson & Walker (a performance and tuning advisory group) puts it: “these are bargains that shouldn't be overlooked”.
SHARE members should, however, also be aware of some of the downsides of the new System z:
- On the downside, while the migration from a z196 to a zEC12 is straightforward (such as a move from a z196-4xx to a zEC12-4xx), it is more difficult moving z196-7xx machines to the zEC12 (because the increase of 25% speed results in increasing the MIPS/CP [central processor] and reducing the number of CPs);
- IBM has intimated that performance for certain applications may increase by up to 48% using the zEC12 — but it may take a lot of effort to see this kind of dramatic improvement (such as recompiles or making use of new instruction sets). Don’t expect your applications to run tremendously faster out-of-the-box — expect that there will be a little effort involved to improve application performance.
- IBM’s new zAware environment sounds very promising (it is a management environment that uses a separate LPAR to collect data from multiple images in a parallel sysplex, which can then be used to streamline major shifts in workloads). This seems like limited functionality for the current $40K/10 CPs price tag. (Expect IBM to do a lot more with this product over time — but until then, think seriously about how to justify this cost for the functionality that it delivers).
Summary Observations
Enterprise IT professionals should look beyond the usual speeds/feeds/capacity improvements to the new System z, and should instead consider the message that IBM is sending with the improvements the company has made to the processor and related on board cache. IBM is saying “we have the most scalable server in the industry with the market’s faster processor. We are positioning this server as a massive on-line analytic processing environment, as a monster transaction engine, and as a system capable of running thousands of Linux images at a cost that is significantly less than deploying an army of x86 servers.”
What IBM is trying to do is shift the mainframe conversation away from a discussion of technologies into a discussion about workload processing. And this is exactly what IBM should be doing in order to overcome the age old objection that the mainframe is an old technology primarily designed for batch and transaction processing.
To join in the conversation about IBM’s zEC12, visit the LinkedIn discussion.