Gene Amdahl, the chief architect of the IBM System/360 (the original, direct lineal ancestor of today’s modern IBM z Systems) and former IBM Fellow, died on November 10. He was 92. If you care at all about computing, pause for a moment to reflect on the passing of (most probably) history’s greatest system architect.

Amdahl’s career at IBM lasted only a total of 13 1/2 years in two stints, but what a stellar career it was (and outside of IBM, too). He worked on the landmark IBM 704, 709, and Stretch computer systems, then, in his second stint with IBM, he helped transform computing forever with the System/360. IBM invested a reported $5 billion (1964 dollars) in the System/360 project, a breathtakingly vast R&D investment to back Amdahl’s architecture. Fortunately the computing architecture Amdahl defined is history’s most durable, and IBM has recouped its investment many times over. That architecture thrives today in incredibly evolved form as the IBM z System with no physical correspondence whatsoever to the System/360 but still with Amdahl’s original design principles at its core. The design foundation was so strong, so enduring, that it’s extremely common for code written in 1965 to be running on today’s latest IBM z13 machine, unmodified, right alongside (and interacting with) 64-bit Java code written ten minutes ago, for example.

Amdahl was named an IBM Fellow in 1965, IBM’s highest honor in its technical and engineering professions. He had a falling out with IBM in 1970 over plans for a supercomputer. (IBM management didn’t think Amdahl’s ideas would be profitable.) After Amdahl left in 1970, IBM embarked on the Future Systems project that, in hindsight at least, was overly ambitious. Parts of the FS project ended up being useful, but for the most part FS was a business investment failure. In contrast, Amdahl founded a company that bore his name that, beginning in 1975, grew to become one of IBM’s biggest competitors ever. Amdahl left his company in 1979, but Amdahl’s “plug-compatible” mainframes reached about 22% marketshare and were still reasonably competitive well into the 1990s when Fujitsu, Amdahl’s initial investor, bought out the company.

IBM bet heavily on several key technologies in the 1990s that were, even in hindsight, risky, expensive bets, including moving from bipolar to CMOS processors throughout the product line, bringing Linux to the IBM mainframe (and with full IBM support), and the development of 64-bit z/Architecture. (The first 64-bit mainframe, the IBM zSeries z900, started shipping in 2000 and was a big success, kicking off the new century’s “mainframe renaissance.”) The CMOS transition was particularly difficult since the first CMOS processors were slower than the older bipolar technology on single threaded tasks, and many workloads are sensitive to single thread performance and throughput. Amdahl (the company) and particularly Hitachi, the other plug-compatible mainframe manufacturer, enjoyed a few years of increased sales during this rocky transition period. However, the writing was on the wall. Amdahl and Hitachi were not able to make the big investments to continue improving their mainframe designs, and z/Architecture in particular sealed their fates. Innovation also accelerated in software, beyond the boundaries of the physical system design. Hitachi and Fujitsu still continue to deliver mainframe systems in their domestic Japanese market with MVS-like domestic Japanese operating systems, though even in Japan IBM’s marketshare has eclipsed them both due to powerful economic forces and the engineering challenges in high-end server development.

It takes a team of great engineers to develop great technology, and Amdahl had a lot of great engineers working with him. That said, by all accounts he was an amazing, inspiring engineer and system architect. Thanks, Gene.

Connor Krukosky, an 18 year old college student, has installed an IBM z890 mainframe in his parents’ basement in Maryland. He posted some photos of his personal mainframe, and at last report (on IBM-MAIN) he has successfully booted Linux on z and can connect. His next task is to get some disk storage attached and working. You can follow his progress on the IBM-MAIN list.

Connor reports that he paid $237 for his mainframe, a capacity model 320 (approximately 120 PCIs according to IBM’s LSPR table). That means that Connor’s z890 has 3 of its 4 main processor cores configured as CPs, and CPs are technically capable of running any workloads, including Linux. It’s possible the 4th core is configured as some other engine type, though that’s unclear at this point. The z890 was available with anywhere from 8GB to 32GB of main memory, so Connor should have plenty of memory to support his personal mainframe workloads. It’s unclear exactly what I/O adapters he has installed in his particular machine, but the z890 was available with 2Gbps FICON/FCP and 10Gbps Ethernet.

IBM introduced the z890 in 2004 and halted new z890 sales in 2007. The z800 was its predecessor, and the z9BC was its successor. One of the reasons Connor likely got a great price on his used z890 is because z/OS 1.13 is the last release of z/OS compatible with this model. There are still some Linux distributions compatible with the z890, however.

Christopher O’Malley, Compuware’s CEO, cites some recent, extremely interesting economic analysis from Dr. Howard Rubin. Rubin studied various businesses to look at IT productivity: how much value are they getting for their IT spending? And how does that value vary depending on where they spend their IT dollars (and euro, and yen, etc.)? In particular, if they emphasize mainframe-based IT, how much “bang for the buck” do they get versus mainframe avoidance?

The answer is that “mainframe heavy” businesses simply get a lot more bang for the buck.

As examples, when Rubin looked at retailers, he found that the mainframe-based companies averaged $194.09 in total IT spending per SKU (stock keeping unit). In contrast, retailers that relied more heavily on distributed servers spent an average of $252.27 per SKU on IT. Among insurance companies, the mainframe-oriented insurers spent $56 on IT per claim, and the distributed server heavy insurers spent $92 on IT per claim. There’s also some evidence that the value gap is widening, and Rubin offers some interesting hypotheses explaining why that might be happening including “starburst effects” associated primarily with fast growing mobile interactions.

O’Malley argues strongly in favor of the “post-modern mainframe.” I like that term. I would interpret the post-modern mainframe (and modern enterprise architecture) as:

  • the centralized (or re-centralized!) enterprise System of Record;
  • also acting as the enterprise security hub providing common security services;
  • still running well maintained, progressively enhanced and encapsulated, concurrent batch and online business logic of whatever vintage(s) that still has value to the business;
  • with additional, popular run-times and expanded cloud capabilities;
  • with additional, on-platform System of Engagement and System of Interaction capabilities (including especially real-time/every-time analytics);
  • closely paired with cloud-based System of Engagement and System of Interaction capabilities where they make sense.

Technically that’s not hard to do, really. Much of this formula you get if you simply stay reasonably current in your software versions. For example, you get the WebSphere Liberty Profile with the latest CICS Transaction Server releases. O’Malley is optimistic that leading companies are abandoning the “hairball” deployment architectures. There’s some evidence of that.