Power7 Blades Plus i Versus X64 Blades Plus Windows
May 24, 2010 Timothy Prickett Morgan
The price/performance saga continues, and in this week’s issue of The Four Hundred, we pit the new Power7-based Power Systems 700, 701, and 702 servers against their rough equivalents in the X64 blade server world. Two weeks ago, when I fixed an error I had discovered with the PS701 blade configurations, I showed you that the Power7 blades were a mixed bag, with the entry i blades able to hold their own against IBM‘s AIX on the same hardware and configured with the Oracle 11g R2 Standard Edition One entry database.
As I moved up to the PS701 and PS702 blades, beefing up the configurations to support more users in a balanced way, then the comparisons with the AIX-Oracle combo got worse and worse, mainly because IBM’s license charges for i 6.1 or i 7.1 are very much out of whack with the pricing for AIX and Oracle. (I used named user configurations for Oracle, which is the closest thing you can do to the per-user charge IBM has for i 6.1 or i 7.1 and its integrated relational database management system.)
As the i-AIX comparison table showed (which you can jump here to see), the i premium is around 10 percent on the PS700, and rises as the machine gets more users, until you are in the 18 to 26 percent range for machines configured with 40 or 150 users. (These are the bands IBM used for user-based pricing with i5/OS. I didn’t just pull 10, 20, 30, 40, 80, 150, and 300 from nowhere in these comparisons.) As you get into the heavier PS701 configurations–this is the single-socket server that can have all of its eight-cores activated and that has 16 memory slots, unlike the PS700, which is a single-socket machine with only half the cores and memory slots–the machines get close to double in price, mainly because IBM seems to think $14,995 per core is a good price for i 6.1 or i 7.1. Based on the AIX-Oracle competition, which costs half to one-third as much on the PS701 and PS702 blades (the latter being the double-wide, two-socket blade using Power7 processors and packing 16 cores in a two-fisted punch), I would say this was a bad assumption for Big Blue to make.
On Friday night, to bring this issue of The Four Hundred on home, I whipped up some new comparisons, pitting the Xeon-based ProLiant BL280c G6 and BL460c G6 blade servers running Microsoft‘s Windows stack and VMware‘s vSphere 4.0 virtualization. This is the comparison that matters as far as the 21st Century midrange is concerned, with Windows being the dominant platform–and by far–among small and medium businesses. And guess what?
The Windows stack on similarly powered Xeon blades–that’s Windows Server 2008 R2 Standard, SQL Server 2008 R2 Standard, and vSphere 4.0 Advanced–is freakily, weirdly, sychronicitously in phase with the AIX-Oracle SEO pricing on the Power7 blades.
It’s disconcerting how close it is, in fact, which you can see in this week’s monster table comparing PS7XX blades running i against ProLiants running Windows. This is particularly true for the PS700 and lighter PS701 configurations, where Windows and AIX seem to be in lockstep. But as the Power and Xeon machines get beefed up, the disparity between all three platforms–i, AIX, and Windows–grows. And that is because Windows is priced per server and incremental processing capacity from Intel is relatively cheap, and so is the price of DDR3 main memory on X64 iron.
In the comparisons I put together, both the Power and Xeon blade configurations include a blade chassis–the BladeCenter S for the IBM boxes and the c3000 chassis for the HP boxes. Both setups have an LTO-3 tape blade (which are crazy expensive), two on-blade disk drives, and external SAS-based disk arrays on blades that slide into chassis. I wanted to look at base hardware and software costs, so I took out maintenance on the hardware and software stack. I had to pick some pretty geared down Xeon chips to get something close to a single-core of Power7 oomph, but I think the comparisons are about as fair as anyone can do. I have tried to reckon what the transactions per minute (TPM) ratings would be for all of the configurations if they ran the TPC-C online transaction processing benchmark test and then converted this to CPW ratings on the HP machines; on the Power-i blades, I went in the other direction, starting with actual or estimated CPW ratings and converting to TPM. The premium for the Power-i setups is based on per user differences, not per TPM differences, and believe me, this is a kinder and gentler comparison. (IBM tuned the hell out of the AIX boxes, showing a ridiculously higher throughput than it can get on the same iron running i. And I don’t think, as I have said many times, this is something inherent in the DB2 for i database relative to Oracle 11g or DB2 V9.7.)
As I said in the prior story in this series, if you have modest CPW and memory needs and you don’t want to lay out a lot of cash, the two-year-old JS12 blade is tough to beat because IBM tossed in the i 6.1 license for free. While the new PS700 blade in a light or modest configuration can meet the JS12 blade, on a per-user basis for a similar number of users, the JS12 wins hands down, user for user. That’s not in these comparisons, but it is important to keep that in mind as you are shopping.
Have fun checking out the table, and I will start doing other comparisons as soon as I can get data on all the new servers out there. There is lots of very cool midrange iron out there, but not all of it is shipping yet.