Technology News Technology News

Another Perspective


On Wednesday, April 26, if the yentas are to be taken seriously, IBM will announce the last mainframe that will use unique zSeries chip technology, a system bearing the code name Pollux.  The system, based on the z9, will be more than a successor to the z890.  It will represent IBM's last chance to consolidate the low end of its mainframe base before launching mainframes based on POWER chips.  Pollox will also give IBM one more chance to push back the barbarians that have been invading the glass house with systems running Unix, Windows, Linux, and OS/400.

IBM's two prior generations of midsize 64-bit mainframes, the z800 and z890, won over many shops that formerly used 31-bit big iron.  But IBM still incurred losses at the low end of the mainframe market, where users defected in order to to gain access to more modern and less costly systems for applications such as web services, email, ERP, and human resources.  Pollox will face similar but more severe challenges, because alternatives to the mainframe have improved much faster than requirements at mainframe shops have grown.  The new target for Pollux (and therefore the new target for non-mainframe solutions) is pretty big; it will include systems with up to 2000 MIPS or about 350 MSU in processing power.  In addition, if a mainframe shop switches platforms, the new equipment will drag along substantial acquisitions of disk subsystems, tape backup, and other equipment.

The case for retaining the mainframe at many shops is based on the efficiency of legacy systems, the resiliency of the mainframe, and the high cost of migration.  For many shops, it's a very strong case.  But for others, the mainframe they see doesn't look efficient; it looks very expensive and that means inefficient.  Alternative systems may provide more than enough resiliency to replace a mainframe.  And the cost of migration, which can sometimes be recovered in just a few years, is as an expense that will probably rise in the future, a point used by advocates of change who urge their companies to just grasp the nettle right now.

IBM is still working out how to best position Pollux, which to a considerable extent boils down to hardware pricing and special acquisition incentives, such as special deals on systems software and middleware.  The problem for IBM is that just about any deal that works will as a retention strategy will also tempt many users spending big bucks on high-end z900 and z990 systems to move to the new platform instead of a z9.

Officially, the data on Pollux IBM has shared with resellers, selected users, and others it considers close and closemouthed friends is sealed under non-disclosure agreements.  But any fool, including us, can guess what the machine will look like based on IBM's recent history and the specifications of the z9.

Like the z890, Pollux will have one processor module, or book, that probably includes a dozen CPU chips.  Up to four of these chips will be available for use as general purpose zSeries engines, much the way they are on the z890.  Another few chips, probably also four in total, will be available to users who want to use them as coprocessors including the IFL supporting Linux, the zAAP supporting Java, and the new zIIP, supporting some database functions.  The other engines will provide systems management technology and, of course, hot sparing in case any live engine burns out.

Like the z890, Pollux will use microcode to govern the speed of its general-purpose processors, reducing the power of a single engine to, we'll guess, as low as 50 MIPS, possibly lower.  Whether IBM actually offers a machine packing only 50 MIPS is an open question.  IBM might insist that users who want the smallest zSeries power must also buy at least one coprocessor, a tie that would boost the cost of a base system.

At the other end of the adjustable power range, Pollux engines will probably run at something over 80 percent of the speed of a z9 engine.  For back-of-the-envelope calculations of capacity, a prospective user might peg an engine at 500 MIPS or 80 MSU.  (The engine rating in MSU for billing purposes is likely to be lower, a stipulation that reduces the cost of MSU-based software licenses.)

Pollux will support more main memory than any user is likely to need, probably 128 GB like the low end of the z9 range.  It will also let users get as many channels as they might want, up to 80 FICON connections and possibly 200 FICON Express links.

On paper, Pollux would be an alternative to z9 systems with up to 3 live zSeries engines and, for some shops that don't run their mainframes at max, even 4 live engines.  And that's what has to concern IBM.  A user who can squeeze into a fat Pollux with only, say, a hundred MIPS to spare, will then be unable to grow much without either adding another box in a sysplex of moving to z9.  That user will probably pay IBM less than the cost Big Blue would want for a similarly powered z9, not only in hardware costs but also in maintenance prices and software license fees.  The better the Pollux deal, the more likely it will be that users of flagship mainframes who don't need more than 2000 MIPS will lean toward the new box and the less IBM will take in from this portion of the target market.

On the other hand, if IBM prices Pollux high, it will increase its exposure to a number of alternatives.

The least costly option for many mainframe shops is likely to be a process based on dual-core AMD server chips or on the single-core Intel X86 server chips that will go dual core later this year.  These platforms will support Linux as well as Sun's Unix plus, of course, Windows.  (Other Unix alternatives are available but are not hot sellers in the Pollux target area.)

The other hardware alternatives that mainframe shops consider when bailing out include the much improved line of processors from Sun, which will soon include 16-core chips that pack quite a processing wallop.  There are also Itanium machines from HP, pSeries boxes from IBM, and i5 servers, also from IBM.  Each of these choices drags along its own selection of storage options, dbms software, other middleware, applications packages, and support programs.

IBM's big competitive headache comes in part from the rapid progress in server technology that has allowed every alternative platform to more or less double its price/performance during the past 18 months and to promise similar increases in efficiency during the coming year or two.  While new mainframes always offer more bang for the buck than their predecessors, IBM has been unable or unwilling to stay with the pack.  It argues, often successfully, that the high tuned workloads already on mainframes can't be moved to other platforms without forcing users to buy vastly more server power and also to go from single-box (or small box count) solutions to populous server farms.  Some users looking at their alternatives say, "So what?" But most are intimidated by the possible complexity of mainframe alternatives and their fears are not always vanquished by vendors' promises of an equalizer in the form of training and support.

All the big questions will be answered in just a few weeks, or at least given preliminary answers.  If IBM gets it right, it will report a deluge of first day orders.  If not, it won't be long before frustrated mainframe customers express themselves.  Even the ones that buy Pollux might insist that the machine comes on castors. 

— Hesh Wiener March 2006

Copyright © 1990-2019 Technology News of America Co Inc.  All rights reserved.