The System iWant, 2010 Edition: Entry Boxes
February 1, 2010 Timothy Prickett Morgan
I am running out of time to pitch the rest of my theoretical and completely hypothetical System iWant, 2010 Edition, machines, with the initial Power7-based Power Systems launch due in February. The current rumor has it that IBM will roll out an entry Power 520-class machine in the next two weeks, but I have obviously not been able to confirm that with Big Blue. What I do know for sure is that IBM has to do a better job with entry Power Systems and that the company knows this.
With the exception of some special packaging of software on entry 520-class machines for i/OS customers, IBM has not really paid that much attention to the entry Power-based servers for a while. IBM’s whole pitch of server consolidation using AIX, Linux, and i/OS both to chase its own customers and get them to upgrade old iron and to go after competitors (mostly Sun Microsystems, now Oracle, and Hewlett-Packard and their respective Solaris and HP-UX platforms) naturally puts Big Blue in the position of peddling midrange and big iron Power machinery. If IBM can get customers to do a Window/X64 or Linux/X64 conversion to one or more of its Power operating systems, it is highly unlikely that pitch would involve a Power 520 box. The Power 520 is a relatively big box, physically speaking, and offers no compelling performance or price advantage compared to Wintel or Lintel iron–or Winpteron or Linpteron boxes, for that matter.
But that doesn’t mean entry systems are no longer important, and that innovation is not required here. Au contraire, mes amis. System vendors that neglect entry products are doomed, since these are the feeder systems for the product line, drawing in customers. Some of those customers–but certainly not all–will grow their businesses, and add more complex software, and if they have a good experience with that entry platform, they will stick with it. A vendor that does not solicit new customers on entry boxes is, in effect, cutting itself off from its broader future.
The AS/400 product line has always been comprised of entry, midrange, and high-end boxes, and the customer base used to be more or less a pyramid. In the past decade, for a lot of complex reasons having mostly to do with money, as IBM has radically increased the performance of its Power-based servers, AS/400 shops didn’t grow their workloads on the i/OS platform quick enough to require very large boxes. And so, these days, the vast majority of i/OS shops today buy an entry Power 520 box and that is more than enough for their database and application needs. In effect, midrange customers stayed in the past, in terms of their data processing requirements, even though they are moving to more modern systems.
I think this needs to change, of course, and I know you do, too. And all the Smart Cube appliance servers in the world are not going to change the fact that IBM needs to not only sell i/OS-based Smart Cubes, but offer upgrades to the regular Power Systems machines and provide the kind of performance, pricing, and features that will convince i/OS shops with their heavy Windows usage to move more workloads over to the Power box. I have been over the Windows conundrum, and I am not going to go through all that again, except to say that the problem is most acute at Power 520 shops using the i/OS operating system.
So what should the Power 520-class machines, which hopefully will get a much more clever name than that, look like? I have a couple of ideas.
First, the Power7 chips are coming off IBM’s 45 nanometer SOI wafer baker lines in East Fishkill, New York. While IBM says that these 45 nanometer processes have ramped nicely since last year and are ahead of schedule, with such a large chip as the eight-core Power7 chip, there are no doubt going to be yield issues. As there always are with any new chip and any new process. (By the way, this is why Intel never does a new chip and a new process at the same time. And which is why IBM probably shouldn’t have, either. You do a crank on an existing design in a new process and when you perfect it and ramp it, you use that on the new chip–what Intel calls the tick and tock of its wafer baking operations.) With the Power7 chips, which have eight cores, a shared L3 cache made out of 32 MB of embedded DRAM, and integrated DDR3 memory controllers and probably a few more features the company has not divulged, there will be plenty of boogers on the wafers that make the chip not suitable in a system that needs all eight cores. But many of those dud chips will work fine with one, two, or four cores and at least one memory controller. And rather than throw them in the scrap bin, IBM should use them to make very, very inexpensive i/OS boxes.
Notice how I didn’t say Power Systems boxes? That was intentional. I want IBM to make Power Systems i boxes with only one Power7 core, maybe running at only 2 GHz to 3 GHz if that is as low as the clock speeds will go in the duds. (That’s maybe 1,500 to 2,300 CPWs of oomph, based on my wild guess.) Throw on enough memory to make this useful, and if it doesn’t support WebSphere going very fast and all its bells and whistles, that’s OK. This box is designed for green-screen workloads with third-party application modernization tools that, with IBM’s help, can be ported to shadow Linux partitions or run on an X64 processor card with Windows plugged into a PCI-Express slot. Whatever it takes to get customers to get current. Now, do the same thing with Power7 chips with only two cores working and give customers a push-pull upgrade between these machines, and then upgrade paths into the real Power Systems lineup, which offers capacity on demand capability to turn Power7 cores on and off as needed and lots more memory, disk, flash, and I/O expansion.
IBM also needs to get away from the idea that one chassis fits all customers. The 4U black chassis that has been the workhorse of the iSeries and System i product line since the Power5 chips came out many years ago needs to get some options–the same kind of rack and tower options that IBM’s System x servers have. That means 1U, 2U, and 4U rack-based servers with equivalent towers, and 7U and maybe 10U chassis for customers who need lots of processing and storage capacity. There is no reason by IBM can’t pack as much storage into is Power7 boxes as Sun has been able to do with its Sparc T and X64 servers. And if IBM allowed storage in the back and front of the chassis, as Super Micro does in its storage arrays, that might be cool, too.
And by the way, it should be possible to change one chassis for another without buying a whole new machine. Why not? Why should changing the case be any harder than changing a coat? If the system was designed to not only upgrade the processors as well as the shells, customers could preserve more of their investments. IBM already standardizes a lot of the internal guts in the Power Systems machines, and it would be a competitive advantage to add more flexibility here. No other server maker does this. You need more peripherals? You need to upgrade your box. Yes, I know IBM has remote I/O drawers in its System x and Power Systems lines, but have you seen the prices it charges for an empty metal box with some wires?
I have a special request for another entry Power Systems i box. I want IBM to build an entry tower configuration that is based solely on flash disk. And then I want to see a decent price on this and then a TPC-C online transaction processing benchmark test result for this machine. Get back out on the front end of technology and performance and start demonstrating that this machine gives great bang for the buck.
Speaking of pricing, I have one piece of advice: Whatever the configuration, whatever the elements of the system, whatever IBM does with the entry Power7 machinery, the pricing of the equivalent Power Systems i edition has to offer better bang for the buck than the Windows equivalent. And it has to prove it. Windows is the enemy down here, not Unix. Let the AIX people worry about Slowlaris and Hockey-PUX, and let them compete as they see fit.