Pondering Possibilities With More Power7+ Machines Impending

February 4, 2013 Timothy Prickett Morgan

With new Power Systems and PureSystems machinery in the works and hopefully set to shake up the market and chase some of those server dollars that Intel is already counting on, now is a good time to stop and have a good THINK. Just like Tom Watson admonished us all to do soon after taking over what became International Business Machines nearly a century ago. If we were sitting in a bar together, we might even think and drink at the same time, as Watson himself did as a young man before he became a teetotaler.

Here’s a funny story, come to think of it, which I had forgotten even after I read Father, Son, & Co, Tom Watson Jr’s biography, when it came out in 1991, just as Big Blue had lost its way and went up onto the financial rocks.

Watson was born in Campbell, New York, up on the southern tier of New York, an area I know pretty well because my own family migrated up north to the region. Watson was a traveling salesman peddling pianos and organs for a number of years–I cannot imagine a tougher job in a tougher place–and eventually moved to Buffalo, where he sold sewing machines for a company called Wheeler and Wilcox. And here is why Tom Watson swore off alcohol, as a cite on Wikipedia lifts from Father, Son, & Co, as retold by his son, the guy who brought you the System/360 mainframe in 1964:

“One day my dad went into a roadside saloon to celebrate a sale and had too much to drink. When the bar closed, he found that his entire rig–horse, buggy, and samples–had been stolen. Wheeler and Wilcox fired him and dunned him for the lost property. Word got around, of course, and it took Dad more than a year to find another steady job.”

Just in case you think the man didn’t at one time enjoy a drink, or he always had it easy. As I write this on a Friday night after a hard day’s work, I am looking forward to a nice, cold, homebrew myself. I suspect you are reading this as you enjoy your morning coffee, and it is probably best you not put anything else in it but milk and sugar at this hour. Unless you have a full day of meetings. . . .

Now, back to Power7+ iron. With the exception of a few thousand large customers, the Power 770+ and Power 780+ machines launched last October using the new processors are way overkill for the vast majority of IBM i shops, who tend to buy modestly powered rack or tower servers to run their back-end applications and the databases that underpin them. So the first thing we all have to be grateful for is the fact that IBM is now putting entry machines at the tail end of the Power Systems launch, as happened with the Power7 chips that came out in February 2010. We didn’t see Power 710 through Power 740 machines until October of that year, and that was a long, long time for the market to wait.

The fact that IBM looks to be launching both the entry and midrange products at the same time is a very good indicator that the yields on Power7+ chips, which are etched using IBM’s 32 nanometer copper/SOI processes, are pretty good. So I am glad that we are getting these machines out the door now and into the market, which will keep it from stalling for months as we all wait to see what IBM will do.

I have been thinking about my wish list for what I want to see in Power7+ entry and midrange systems for quite some time. I think there are a number of things that Big Blue can do to strengthen the product line to do battle better with X86 machinery and also add in some extras that will help please its existing IBM i and AIX installed bases. Officially, in this newsletter what we care most about is IBM i, but anything that makes Power Systems stronger makes IBM i stronger and live longer. So we also root for AIX and Linux on Power–and have also advocated for Windows and MacOS on Power, too, as well as for PHP, Perl, Python, Ruby, and just about anything else that a modern computer might run.

The first thing I would like to see with the new Power7+ entry and midrange machines is for IBM to really take the fight directly to the X86 platform and to realize that Power, not X86, is the underdog for the kind of complex distributed workloads that companies are deploying on grids and clouds these days. IBM has to meet the two-socket X86 workhorse server on its turf both in terms of price and performance, and I would argue that it has to beat them on both of those fronts. If IBM can’t do that with Power Systems and its PureSystems derivatives, then it has a real problem. If IBM has to charge more for the hardware, then it will have to charge less for the systems and management software than runs on top of it. As it is, Microsoft, Red Hat, and VMware are getting more profits for their operating system, virtualization, and management tools than any of the server makers can get out of bending metal and reselling chips.

IBM has to find a way of getting the cost down on these machines, and hopefully it has done that. (This will be made somewhat more difficult for IBM Microelectronics if Sony shifts away from the Cell PowerPC to a mix of CPUs and GPUs from Advanced Micro Devices, as has been rumored. IBM’s embedded Power chip business makes it possible for the company to make its own chips. So do the mainframe and the Power chip businesses, and IBM could always do something interesting and buy AMD and become an ARM server chip supplier as well as an X86 supplier in its own right. IBM could, for instance, buy AMD and sell its fabs, call it a wash. That would be fun, for sure, but it is not going to happen unless that Sony contract and others really matter more than we think.

There is only a brief window of opportunity before the next-generation “Ivy Bridge-EP” Xeon E5 v2 and “Ivy Bridge-EX” Xeon E7 v2 processors come to market in the second half of this year, and IBM needs to be able to compete, and compete hard. And the Xeon v3 chips, code-named “Haswell,” are going to be even more impressive. Intel has the process advantage, building chips in a 22 nanometer process that is fully ramped, but the Power instruction set and feature set can still be shown to have performance advantages. Hopefully it is enough.

IBM needs to remove the cost differences between X86 systems and Power Systems off the table. I just don’t know if it can make any money doing that.

Here are some other things I would like to see IBM do in conjunction with this new Power7+ iron that is coming out:

i5/OS V5R4 Emulation Environment. This is not a new idea, but IBM needs to use that QuickTransit emulator and help customers who cannot move their code over to IBM i 7.1 running atop either Power7 or Power7+ chips. I know that many customers are balking at this because they don’t want to pay ISVs back maintenance and other fees to get their code current if they use third party software. That said, there are others who have homegrown applications that would be happy to pay a performance penalty if they could just drop a V5R4 runtime onto a logical partition and run the apps in emulation mode until they can gradually port the apps over to run native on Power7/7+ plus IBM i 7.1. The base of users is not just the ones the ISVs can see, but also the ones that IBM, resellers, and ISVs can’t see.
Workload consolidation bundles. This one is easy enough. If you are attacking X86 workloads with Power Systems, you can’t just take this naked iron and throw it into a data center. It is always about the software. You need workloads that you can move off Windows and Linux iron and onto Power Systems machines. IBM is talking about starting up an Exchange-to-Domino migration service again, eventually supplying tools. But I think what IBM needs to do is create a stack of software that runs natively on Power iron that can actually replace those X86 workloads. I realize that the mixed-mode PureSystem family is basically an admission that this kind of workload consolidation can’t be done. I don’t believe that. For instance, imagine if IBM went to Google and licensed its backend systems for mail, chat, office automation, and other common tasks that we use online and embedded them into Power Systems operating systems like IBM i and AIX. This would probably infuriate the former Lotus division. I don’t care. I am interested in selling Power Systems here.
Specialty engines. The mainframe has them to goose the performance of database and Java workloads, why doesn’t the Power Systems lineup? These specialty engines would fit hand in glove with workload consolidation, just as it does with the System z mainframe. In the past quarter, IBM’s MIPS shipments for System z mainframes were up 66 percent, but half of the MIPS shipped were for engines running Linux or zIIPs to accelerate DB2 or zAAPs to accelerate Java. These engines and their software cost a lot less money than regular mainframe engines, so I think the same should hold here, too. I think IBM could carve out cores on Power Systems and, for instance, install the Zing turbo-charged JVM from Azul Systems to radically boost Java. The idea is not to make Java on Power Systems run faster, per se, but to get Java workloads currently running on Windows and Linux machines based on X86 architecture onto the Power boxes. You could similarly work on accelerators for PHP, big data munching and queries, and other workloads. You could even have accelerators that just speed up transaction processing in DB2 for i, so rather than buying a whole license for IBM i for a core you might instead just buy one that allows for a primary core to offload transaction processing work to it for less money. The idea in each of these cases is to make it up in volume, to grow the base.
PureSystems HA clusters. I have always thought that high availability should be transparent to customers, particularly in an integrated machine like IBM i. If I were IBM, I would build a fully redundant machine with HA software, partnering with Maxava and Vision Solutions to create an HA system that automagically replicates data and applications and is so good that it fails over automatically and no one even notices. (Hey, I am dreaming here, so we might as well dream big.)
Fault tolerant Power Systems. How come Hewlett-Packard with Tandem, Stratus Technology with its ftServer, and NEC with its mirrored Express5800s are the only vendors with true fault tolerant hardware? Where is the Power-FT box? Give the HA software partners a piece of the action and do hardware mirroring. This may be more work than the money it generates.

More than anything else, what I really want is for the Power 710+ through Power 760+ machines to compete, head-to-head and toe-to-toe, with X86 machines. The future of the Power Systems market depends on it, and so do we all.

I am just glad IBM didn’t port OS/400 to Itanium.

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search