Happy New Mainframe Day! Introducing the zEnterprise EC12

Good morning in the Americas, and happy New Mainframe Day! Today IBM is announcing the zEnterprise EC12. You can order one (or more) today, and they start shipping in about three weeks. (IBM is just like Apple that way: no vaporware.) And the world's flagship enterprise server is practically perfect in every way.

I'm still reading through the wealth of information that IBM has revealed today. Every major general press outlet is reporting on this announcement. Here are my quick takeaways:

  • Another giant leap higher in processor clock speed, and another world record. Wow, I am in awe. IBM people tell me they expect to maintain industry clock speed leadership "indefinitely."
  • Hexacore architecture with about double the number of transistors per die compared to the z196. That's a wonderfully nice increase — but it amazes me that IBM's engineers figured out how to run and cool 6 cores per die at mainframe service levels and at such incredible, continuous speeds.
  • About 25% performance improvement per core and 50% performance capacity improvement per server. Those are big jumps. Has anybody else noticed that's with only an amazingly short 25 month product cycle since the z196 debuted?
  • Cache is up big, although the cache mix across levels is slightly different. Where you "spend" your cache real estate matters, so apparently IBM found a better arrangement for spending that much bigger cache budget.
  • New "server class memory," i.e. directly addressable flash storage, another industry first-and-only. Including this server class memory you can have about 9.4 TB of directly addressable memory per machine -- all redundantly protected (RAIM), another industry first-and-only. Right out of the gate z/OS exploits this storage class memory for things like paging, memory dumps, etc. for faster problem determination and recovery. More exploitation to come, of course.
  • New processor instructions and loads of new software that exploits those new, faster instructions, with a particular emphasis on information analytics.
  • zEnterprise is IBM's first and most advanced hybrid computing server, and now IBM has upgraded the zBX to a Model 003 along with the zEC12.
  • A new "zAware" option (which I need to read more about).
  • Up to 101 customer configurable cores per server, up from 80. (N.B. There are scores or even hundreds of other processor cores inside such a zEnterprise server, even excluding the zBX option. Be careful if you count and compare.)
  • IBM held the line on space/power/cooling requirements. Bravo. They've also added a new, even more redundant/more reliable cooling system.

I will probably post some updates a bit later but, in the meantime, let's read on together.

UPDATE #1: I've corrected the total memory figure above. Computerworld has an interesting story about the new zEC12, including these additional highlights:

  • The storage class flash memory is encrypted, presumably because conceivably someone could physically get to it (unlike zEnterprise DRAM). IBM does think of these things. Moreover, IBM zEnterprise CTO Jeff Frey explains that IBM plans DB2 and Java exploitation of this new storage class memory. One example: in the not-too-distant future DB2 will be able to keep extremely large databases in directly addressable nonvolatile memory. Now that would be very interesting and useful.
  • As usual, some workloads do better than others when IBM increases the performance of its mainframe, although all should do very well. Some of the ones that should do particularly well are big multithreaded Java applications, compute-intensive C/C++ code, DB2 (especially in analytics), and SAP.
  • While IBM's z114 (and predecessors) have had the option to install without a raised data center floor, now that's available for the big zEC12 too. Eliminating the raised floor makes it easier to install a zEC12 in a crate, I would point out, so you can have interesting portable data centers.

The IBM announcement also mentions the fact that the zEC12 is the first commercial server to include processor instructions supporting transactional memory, not counting IBM's unique supercomputer for Lawrence Livermore Laboratory. Now, I have a bone to pick here: the computer science engineers (presumably) haven't given this feature a good name. The word "memory" is misleading, in my view. But so they've decreed, so that's the name. The new transactional (blank) instructions make it easier and faster to program with concurrency control. And that's a very good thing.

UPDATE #2: Mainframe Watch Belgium has much more information, including these highlights:

  • IBM has introduced 2GB page sizes and supporting instructions in the zEC12, which are great for DB2 and Java, in particular. So now that's 4K, 1MB, and 2GB page sizes. Considering that we've had only 4K forever, and 1MB debuted only recently (in the System z10 in 2008), this is amazingly rapid evolution.
  • IBM is going to eliminate zAAPs in favor of zIIPs. Initially with the zEC12 IBM will provide z/OS PTFs (enhancements) for running all zAAP-eligible workloads on zIIPs, even when zAAPs are installed. But the zEC12 is the last machine to support zAAPs. Translation: your capacity planning just got simpler. I expected this at some point, but I didn't it expect it yet. It's great news.
  • There's a significant software technology dividend across IBM software. Or, said another way, you can do more with less money. Nothing wrong with that.
  • The machine looks really spiffy, even artistic. No, it's not pink or purple, but perhaps IBM will entertain even that request if you want.

UPDATE #3: IBM's Nick Sardino takes us on a guided tour of the new zEnterprise EC12:

by Timothy Sipples August 28, 2012 in Future, Innovation, Systems Technology
Permalink | Comments (0) | TrackBack (0)

Next Generation zEnterprise: Smart, Secure, Efficient

Here's part of the argument for high-end enterprise servers:

UPDATE: IBM is promoting a major announcement on its zEnterprise home page. To register for the webcast visit this site:


by Timothy Sipples August 21, 2012 in Future, Innovation
Permalink | Comments (0) | TrackBack (0)

Musings on the Evolution of the Server Market

I've been doing a lot of thinking lately about trends in the server marketplace, in part based on conversations with server buyers but also as a student of history. To summarize, I think the world has decided there should be three basic types of servers:

  • High-end servers (aliases: enterprise servers, mission-critical servers, vertically scalable servers, large SMP servers, "system of record" servers, "data center in a box"). These servers are becoming more important every year as businesses and governments cope with massive amounts of data, as security needs grow, and as the consequences and costs of business process failures rise. Preeminent among these servers are IBM's zEnterprise machines, although IBM's high-end Power servers are also firmly in this category.
  • Workload-specific servers (aliases: appliances/appliance servers, application-specific servers, workload-optimized servers). These servers generally run one or a few highly related workloads that are packaged together and, at least to some degree, tweaked and tuned. They often include integrated disk and/or solid stage storage. There's a range of capabilities in this category, from the highly flexible IBM PureSystems to more closed offerings such as Oracle Exadata as examples.
  • Commodity servers (aliases: X86 servers, volume servers, horizontally scalable servers). Obviously there are lots of these servers sold every year, and some buyers, such as big Internet companies, design their own hardware to their own specifications, bypassing vendors such as HP and Dell. Vendors such as Lenovo will also put pressure on HP and Dell in particular. IBM and Cisco have been trying to differentiate themselves with unique capabilities in this crowded market segment. Intel dominates this part of the server market, although it's still an open question whether ARM and/or AMD will be able to make further inroads as CMOS microprocessor technologies reach their natural limits.

There is some blurring between these categories, notably in the small but important supercomputer market which blends characteristics of workload-specific and commodity servers.

IBM has been carrying the high-end torch for years, building the world's most impressive, capable, and innovative enterprise servers. Consequently IBM always has a target on its back — it has always been thus. After all, if you don't have anything to compete, then the only option is to try to dismiss the category entirely. Also, while one of the defining characteristics of a high-end enterprise server is its general purpose nature — the ability to run multiple disparate workloads with varying service level requirements even within one operating system image — I would expect IBM to continue improving the "consumability" of its servers. An excellent recent example is the IBM DB2 Analytics Accelerator which combines zEnterprise, DB2 for z/OS, and Netezza technologies into a single, integrated, extremely high performance information system for both transactional and business intelligence/data warehousing. It "just works." In such ways these servers are workload-optimized while still retaining the full, open flexibility that general purpose high-end servers also have.

As I've mentioned before, the server business is tough. In some ways it resembles the global commercial aircraft industry which is highly competitive but also highly concentrated with only two major manufacturers: Boeing and Airbus. In the server market it's much the same now: IBM and Intel, perhaps with Oracle lately playing the role of Bombardier or Embraer (regional aircraft manufacturers for niche roles) — an even more volatile market segment. There's a simple reason: server (and processor) research and development is extremely expensive. Both IBM and Intel have built business models that, while different, work for them. IBM's server business model is analogous to Apple's, with lots of vertical integration, high value added, revenue stability and predictability, and a broad range of in-house capabilities. Intel's is more like Microsoft's, still its closest partner as it happens.

I am expecting the server market to continue along these three basic tracks for some time to come. I'm also expecting IBM and Intel to continue leading the industry. The vast majority of businesses and governments need some of what both these companies produce.

by Timothy Sipples August 21, 2012 in Cloud Computing, Systems Technology
Permalink | Comments (0) | TrackBack (0)

Oracle Loses Court Case v. HP, Will Appeal

Judge James Kleinberg has ordered Oracle to continue porting some of its software products to HP Itanium servers for as long as HP cares to build the machines. The Oracle software products include those that existed on HP Itanium servers as of September, 2010. The judge even ordered that the products must be complete, fully tested, tuned, supported, and released at the same time as other platforms — which is hard even when you want to deliver something. Tuning is especially different on Itanium.

In a previous post I wondered whether any court would order Oracle to continue doing something it clearly didn't want to do, doubting whether such a remedy would be viable. Well, apparently a judge thinks this particular remedy is operable. We'll see.

Naturally Oracle is appealing the decision.

Would you buy such a software and server combination?

by Timothy Sipples August 2, 2012 in Systems Technology
Permalink | Comments (0) | TrackBack (0)

The postings on this site are our own and don’t necessarily represent the positions, strategies or opinions of our employers.
© Copyright 2005 the respective authors of the Mainframe Weblog.