Research tools - laptops for the Internet and handhelds for mobile usability testing

isham research

The IBM System z9 - the latest z900 midlife kicker

Until z6 becomes available, z990 mid-life kickers and derivatives will have to support the market. The IBM System z9 takes the system to about 600 MIPS/engine - with General Availability in mid-September, possibly the 16th. There are a number of obvious improvements - a reduced cycle time, 40% more processsors, more efficient I/O via a new form of channel indirect addressing, microcode to permit all books to share a single pair of spare processors and fully populated processor chips. The usual stuff of mid-life kickers.

But there is one significant novelty - Concurrent Book Removal. This allows a system with sufficient spare processors (unassigned and unpaid for processors count as spares) to quiesce an entire book so that it can be removed and/or replaced in flight. Certainly within the mainframe world this is the first time that processors can be hot-swapped not only without an outage but also - if the system is properly configured - without a performance impact.

Hot-swapping processors has been done before - but not with this level of virtualisation.

"Properly configured" becomes an interesting concept. With the z990 most users don't know (or care) how many of the unassigned processors are defective - perhaps replaced in-flight by spares. With Concurrent Book Removal, the number of non-functional unassigned processors becomes of interest.

It really is worth wondering why IBM does these orgasmic announcements. "We're doing a large systems announcement in July 05; anyone else want to dump anything into the RFA as it passes?" - and dozens of people empty their desk drawers of - on the scale of things - meaningless dross.

The result is that major selling points - like the ability to remove a large chunk of a running machine for repair or upgrade without even a performance degradation - get buried among features of much lesser import such as real memory size and granularity, 336 FICONs, L1 and L2 cache sizes, etc., etc.

Who cares? Each and every time the reader sees a "feeds and speeds" number (such as the above 336 FICONs) the thoughts are: "Hmmm. Is that really enough? What happens if they fail to make that many work? How many do I need to buy?" Most of this trivia is not just not important - it's not even interesting. One should be able to assume that a system with x times the power of its predecessor (up to 1.95 times, comparing a 54-way System z9 to a 32-way z990) also provides x times the support resources.

The old approach was far better. Announce the machine, let the cycle time and memory issues fade in the public consciousness, and then announce the killer feature - in the System z9's case, Concurrent Book Removal. A real and tangible reason to consider swapping your relatively new z990 for a System z9.

It's a lousy name, by the way. Grammatically interpreted, it means the ability to remove two or more books at once. How about "Transparent Book Removal"? Or "Non-disruptive Book Removal"? Why mention "books" at all - it's a level of archania that doesn't interest CIOs. A throwback to the days when every feature had to be sold as a benefit. "Removal" is also a word with more negative connotations than positive - it hardly implies the benefit that it represents. "Advanced Availability" could be anything - quick-release catches on the doors? But a system that can be repaired in flight without a performance hit - that's something CIOs can understand. And appreciate, if they're the victim of a Service Level Agreement.

Tandem once did it in one word - they called their system "NonStop". Something with the same impact is sorely needed here.

Performance remains a bone of contention. Once again, IBM's Large Systems Performance Reference has been changed between generations rendering many comparisons with earlier machines complex, time-consuming and sometimes impossible.

IBM should finally realise that it isn't fighting Top Gun wars with Amdahl and Hitachi any more, and correction of a single slightly mis-sold machine is a simple commercial issue affecting only one customer. Getting the ultimate single figure out of a system by massaging benchmarks is no longer what the game is about. Raising the ceiling by lowering the floor?

Almost all temptation to use the word "holistic" on this page has been resisted.

And for a weird one - the "Technical Introduction" manual (SG24-6669-00) has become a repository of all sorts of archania, continuing a trend begun with the z900. But what does the following Q&A item have to do with the z9?:

Q: Will z/OS 1.7 run on MP3000 systems?
A: No. z/OS 1.7 requires Architecture Level Set 3, commonly known as "64-bit operation."

Huh? Where does that belong? Do Multiprise customers now have to read z9 documentation?

Copyright © isham research 2005. May be reproduced in part or whole if attribution (preferably in URL form) is given. If used in presentations, a reference in handouts suffices. If used on web pages, a link should be provided. No charge is made. This also applies to IBMers.

isham researchHome Page