Saturday, March 1, 2008

Thirty years on - a new generation!

I had only been back in Sydney for just over a year when IBM launched the “Glendale Series” of mainframes – the IBM 4331 and IBM 4341. The year, 1979, was a very exciting time for IT in Australia as the federal government was allowing a capital investment allowance of 40 percent, as an incentive to industry, which included new computer purchases. However, this great deal was due to expire on June 30th, 1979.

So, with the New Year, corporations were waiting for the announcement of this new mainframe and were anxiously standing by their fax machines eager with anticipation. As January 30th dawned on the US East Coast (January 31st in Sydney), the news finally broke and the price points were stunning. It was a circus! IBM worldwide quickly became overwhelmed with orders and introduced the first lottery – allocating new machines according to a random draw.

These new mainframes featured high-density logic chips, with up to 704 circuits per chip. And with the older IBM 370/158 rated as a 1.0 “old style” MIPS machine (based on IBM’s figures of a cycle time of 115 nanoseconds, which is about 8.7 MHz.), this new one, the 4331 mainframe, rated at about .3 MIP, and its big brother, the 4341, at a shade under 1 MIP – an incredible price/performance achievement at the time.

I was reminded of this as I was in Orlando, FL at an IBM user event to see the launch of IBM’s tenth generation mainframe – the IBM System z10 Enterprise Class (z10EC), and I was anticipating its launch in much the same way as I had some thirty years before. As the events of Monday wound down, and the trade-show floor emptied, IBM engineers descended on the z9 mainframe that was on the stand supporting attendee labs, unplugged it from the storage, and after 4 hours of frantic work replaced it with the new z10. By midnight, it was running, and as the attendees returned for early morning labs they found themselves running on the new mainframe. Whereas the other four sites around the world, where IBM held its press conferences, had only z10EC shells - it was at the user event where a fully operational z10EC made its public debut!

So much has changed since the Glendale Series rolled out, but the performance improvements provide the biggest contrast and with the z10EC, it’s all about the new z10 Processor Chip and the z10EC Multi-Chip Module (MCM). IBM is leveraging the POWER6 Dual-Core chip – no, it’s not the same chip (CISC for z10EC Processor Chip), as its not s a RISC chip as you will find in other IBM offerings – with IBM suggesting that they are “siblings, (but) not identical twins” as there is small number of elements common to both.

Instead of the high density logic chips, with 704 circuits, that we saw on the 4331 and 4341 mainframes, we now have ceramic MCM blocks (103 ceramic layers) each with 5 Processor Units (PU) chips, where each PU had four cores, and where there were now 994 million transistors per PU chip – somewhere near 5 billion transistors associated just with the PUs on the MCM block. Throw in the two Storage Control (SC) chips that add an additional 1.6 billion transistors, and you end up with something like 7 billion transistors per MCM block. A fully populated z10 with four “Books”, IBM-speak for very fat blades, each with one of these MCM blocks, gives you a total of 64 PUs that are capable of processing some 30,000+ old-style MIPS.

Putting this into context, my rough, back-of-the envelope calculation, suggests that just one of these mainframes now has more MIPS than was shipped across the entire life of the Glendale Series. All in a machine that’s not much bigger than the earlier 4341 stood upright! Certainly not the behemoth many of us had grown accustomed to seeing holding court in the middle of the data center. And did I mention that the contrasting color flash down the side of this new mainframe has been changed from red to green – a stark reminder of everyone’s concerns over energy requirements.

But in the press reports that came out on the day of the announcement, there were still some concerns about IBM’s continued investment in mainframes. According to a report by Associated Press “analysts said IBM's advances in chip technology and software are helping the mainframe stay competitive against lower-cost competitors. But they caution that because of price IBM still faces challenges in luring in new customers.” And in the same story, they quoted Brad Day, a Forrester Research Vice President, as having said "this is definitely not a slam dunk - the math still has to be there. The life-cycle-cost-of-ownership argument still has to be there."

The picture I have included here is of me beside the working ”internals” of the z10EC. What you may not pick up all that easily are the very fine optic cables coming down to the multiplexers that support all the external storage. These are running InfiniBand – a first for IBM. When I was talking to an HP NonStop engineering manager recently, and we were talking about InfiniBand, he talked openly about how “the ServerNet footprint is all over InfiniBand.” While IBM is only using it to connect to I/O subsystems, as well as supporting its own Parallel Sysplex interconnect option, I found it highly encouraging for IBM to now deploy a technology with which all of us working on NonStop are familiar.

IBM continues to make extensive use of virtualization – the concept of logical partitioning, or LPARs, is the first layer of abstraction above the metal and is based on earlier releases of VM. “LPARs, on (the z10EC are) not really virtualization with overhead, because it is truly built into the hardware,” suggested a colleague attending the show and helping me sort through all the terminology. In an LPAR you can run IBM’s zVM hypervisor with its support for any number of guest OS’s including zOS and zLinux. While you may want to configure a “relatively small number of LPARs, compared to the number of VM guests,” he added, “there is a significant advantages when running as a VM guest.” IBM’s support of virtualization remains a very powerful feature of the z10EC and is a technology where I do anticipate hearing more from HP, including HP NonStop, in the future.

The timing of my trip to Orlando, coming as it did back-to-back with my trip to South Africa for SATUG, where HP and Intel had provided updates to their product roadmaps, I couldn’t help but notice many similarities, and how the technologies of HP and IBM appear headed in very similar directions. I know that this statement might surprise some of my IBM colleagues – but the more I looked at the z10EC and the more questions I asked, the more I saw how the “top-of-the-line” offerings from both vendors were beginning to look alike.

And has IBM done enough with the z10EC to threaten the HP NonStop marketplace? My first reaction - I am not seeing it just yet! Even in the presentations made by IBM, the proposed growth appears to be organic, coming from upgrading existing customers. Yes, IBM talks about potential future products from ACI (I heard this mentioned a couple of times) as helping, but any movement on that front is years away. I continue to be amazed how the NonStop technology continues to outlive the technologists that propose a change!

For me, the biggest difference comes with the decision taken by HP to standardize on single chip architecture, and to ride that vendor’s roadmap. IBM still uses different chip architectures and while there’s some shared portions, it still requires two different design groups and two separate fab facilities. I don’t get that, and I can’t see that being a sustainable model for the long haul!

The NonStop architecture may be better positioned than the IBM mainframe as HP looks to position NonStop as a “configuration option” on future HP bladed architecture products. In my recent presentations I have simply been stating: “rather than asking the question ‘who will be using NonStop’ perhaps the more appropriate question may be ‘when do I take advantage of the NonStop?’” There will always be a subset of transactions that will be viewed as mission critical, and routing them to mission critical applications deployed on bullet-proof NonStop servers, at a much lower price point, will continue to give HP an edge.

I realize that this blog posting is a little more technical than I have written in the past, and I trust you will indulge me on this occasion. But has the arrival of z10EC changed my view on any of this or become a disruptive technology that will force HP to change its roadmap – well, not exactly! Without taking anything away from IBM – it’s a great system with some remarkable technology that will be welcomed by many of IBM’s customers – it’s not a must-have “killer system”! And I just have to wonder if the green stripe down the side of the box is a slight tinge of envy as much as a testament to its reduced environmental impact!

1 comment:

Dr. Steve Guendert said...

Please allow me to play devil's advocate Richard, but somehow I doubt that the green stripe on the z10 has anything at all to do with envy on IBM's part. I think it may actually symbolize the envy for the mainframe coming from our friends at HP. The overwhelming majority of the world's mission critical (run your business) data still is on mainframes and mainframe attached storage. HP has realized this over the past 18 months and embarked on an ambitious strategy designed to gain marketshare in mainframe storage via their rebranded HDS DASD arrays. I have seen HP hire mainframe skilled storage people, train many people, write a great deal of documentation, and actually seriously compete for FICON director business. In the past they seemed to ignore it. Perhaps they have seen the tide shifting back to Big Iron?