Windows Loses to Power 720-IBM i Combo, But Whips Power 750s
November 1, 2010 Timothy Prickett Morgan
It’s probably a good thing that most of you out there in AS/400 Land are looking to acquire a Power 720 server, and one with not too many processor cores activated running the IBM i operating system. For you, whether you are comparing the Power 720-IBM i tag team to Power-based servers running AIX and Oracle databases, the Power 720 is an absolutely competitive platform, user for user, dollar for dollar.
This would seem to indicate someone at IBM is listening. Which is good news. However, the Power 750, the next stop up for most IBM i shops, has IBM i software prices that are far out of whack with reality, and that will not only be a caution to many Power 550 customers, but also to Power 520 and 720 customers who are worried about that next step down because it is a doozie. This shows that IBM is not listening. I know I have been complaining about this for years.
For most IBM i shops, there are really only four servers that matter. The Power 720 and 750 for rack or tower configurations and the PS700 and PS701 for blade setups. I gave you the lowdown on how the Power7 blade servers, which came to market first with the Power 750, stacked up against their predecessors here, and then eventually got around to showing you how the PS7XX blades compared to Unix versions of Power7 machines. I also showed you how the PS 7XX blades stacked up compared to entry X64 blades running Windows Server 2008 and SQL Server right here. You got the scoop on how the Power 750s stacked up against their predecessors in the OS/400 and IBM family back in August, and I created similar comparisons between the Power 720 and its predecessors a month ago. Last week, I showed you how the Power 720s and 750s running IBM i rate versus the AIX-Oracle combo.
All four machines that the IBM i base really cares about–the PS700, PS701, Power 720, and Power 750–show pretty good bang for the buck against their predecessors, but as I said last week, these machines are not just competing with history, but also with current events in the server racket. And as was the case with Unix comparisons, the smaller PS7XX and Power 720 configurations are absolutely competitive with Windows alternatives, but fatter PS7XX blades and Power 720 setups and all Power 750 machines are priced out of kilter compared to the midrange market they claim to be in. To put it simply, IBM is pricing the IBM i operating system and integrated DB2 database as if it were an enterprise edition release of an operating system and database when, sadly, many of the features that are present in AIX Enterprise Edition, Windows Datacenter Edition, SQL Server Enterprise Edition, and Oracle 11g Enterprise Edition are not in the IBM i-DB2 combo. IBM is also charging too much for 5250 green screen capacity on the Power 701 and 702 blades and the Power 750s.
To give you a broader sense of what raw processing capacity (just processor feature cards and core activations), I walked you through how each feature in the entire Power7 server lineup for which IBM has provided a Commercial Processing Workload (CPW) benchmark test rating stacks up with all the others in the Power7 product line. Then, the following week, I burned the raw processing cards with i 7.1 operating system and Software Maintenance for the processor cores activated on the processor features. (These two comparisons did not include the cost of a server or rack chassis, memory, disks, or other peripherals, unlike the other comparisons.) There is not enough time in the week to do representative configurations of all possible Power7 setups with equivalent X64-Windows servers, so in this week’s patent-pending TPM monster price/performance table, I am just looking at representative Power 720 and Power 750 entry configurations–the ones that IBM i shops would be most likely to buy–and seeing how they compare to equivalently geared down X64 machines from Hewlett-Packard.
The first observation I want to share with you is that in many cases, the Ultrium LTO-5 tape drive that HP is peddling alongside the ProLiant servers is not only a lot more expensive than the similar drive that IBM is peddling in the midrange, but it is a bit more expensive than the entry configurations (with proper memory, disk, and RAID 5 controllers) of ProLiant DL380 G6 or G7 server itself. Yeah, that’s nuts. Welcome to the volume server business and the low-volume Ultrium tape drive and Power Systems racket.
The other observation I will make, and this will come as no surprise to you, is that software makes up a very large portion of the configuration prices shown in the table. (I have purposefully omitted hardware and software support costs in these comparisons to get to the heart of the basic system price.) And no matter what the platform is, the per-user fees that IBM, Oracle, and Microsoft charge for the entry products (usually with a Standard Edition moniker stuck on them) very quickly outgrow the underlying cost of the hardware as users are added to any particular machine. That’s why my comparisons have only a couple of configurations, but a wide variety of user counts.
I think that seat counts, once you get two machines in roughly the same performance tier, are more important for OLTP jobs anyway. This is, in fact, what companies pay for and that means it is, perhaps, as important at looking at cost per transaction per minute. Customers need to worry about peak database performance during the bulk processing they do at the end of a day, week, month, and year, and they need to have some excess capacity for unplanned peaks in the OLTP system. But they sure do not run their systems full out, 24×7. They buy a machine and add users until it chokes and they need to upgrade.
I picked HP ProLiant servers in the DL380 family to compare to the Power 720 and 750 boxes because these are the workhorses of the X64 server market. They offer the same scale of capacity as the 720s and the entry 750s that are most likely to be acquired by IBM i shops. I could have also configured Xeon 7500 machines, but if you are down in this performance area–somewhere between 6,000 and 25,000 CPWs in AS/400 Land–there is absolutely no need whatsoever to move to a much more expensive Xeon 7500 box unless you plan to scale up your machines fast or you need the larger caches and main memories these six-core and eight-core processors offer. Otherwise, two-socket servers with Xeon 5500 and 5600 processors with two, four, or six cores each will do the trick just fine–and give you significantly higher clock speeds to push those batch jobs faster and harder.
To do a comparison of throughout as well as seat counts, I have reckoned the OLTP throughput on the TPC-C test for a given machine based on actual TPC-C test results and then worked backward to give these machines a hypothetical Commercial Performance Workload (CPW) performance rating if you like thinking in CPWs. If IBM still did TPC-C tests on the Power Systems running IBM i and the X64 server makers did a lot more tests than they do–or heaven forbid did their own relative performance metrics to help us reckon TPC-C and other benchmark throughput in a rough way–I wouldn’t have to do this back-of-the-envelope guesstimating. But someone has to do it, and no one else is. So it looks like it is me, again.
As you can see from this chart, the story with the Power 720 configured with IBM i against ProLiants running Windows Server 2008 R2 and SQL Server 2008 R2 is much the same as the story was with Power 720s configured with AIX 7.1 Standard Edition and Oracle 11g Standard Edition. Configurations of the Power 720 with one core activated and a certain number of users can compete against a two-core Xeon box with the Windows stack and VMware’s ESX Server 4.1 Advanced Edition hypervisor.
In fact–and I checked this three times to make sure–on a per user basis for five, 10, 20, and 25 users, the IBM i machine is actually a little cheaper than the Windows machine.
But Microsoft charges for the Standard Editions of Windows Server and SQL Server on a per system basis–that’s $1,089 for the OS with five users and $898 for the database–with just under $40 per user for Windows access and $164 for SQL Server access. IBM charges $2,245 for IBM i 7.1 per core on the entry Power 720s plus $250 per user. Very quickly, as you add cores, the IBM i stack starts to be a lot more expensive than the Windows stack when you get up to 40, 80, or 150 users. Larger Power 720 machines are in a higher software tier where an IBM i license costs $10,995 per core (plus $4,000 for Software Maintenance per core), and the Power 730, 740, and 750 have IBM i fees of $40,000 per core (plus the same $4,000 per core for SWMA). On the bigger boxes, IBM i is even more expensive, at $53,000 per core and $6,000 per year for SWMA.
The next result of this is that the Power 750 running IBM i boxes is just not competitive with Windows-based X64 servers set up for virtualization and database processing. And adding in interactive 5250 processing capacity for the Power 750s makes the comparisons even worse. At least 5250 capacity is built into the entry IBM i licenses. But IBM charges per user fees to access the system, which takes the money right back in a way. Even with the absence of per-user IBM i fees, the per-core pricing and just puts the cost of the Power 750 into the stratosphere:
Not competing well against a Unix machine running the Oracle database is an issue for any platform playing in the midrange, including Power Systems running IBM i. But looking competitive for entry customers and then highjacking their budgets further down the road is unacceptable and positively uncompetitive–and can only lead to sorrow. And not competing well against the Windows stack–at every point in the Power Systems product line, particularly for IBM i shops–is absolutely foolish. OS/400 shops knew for years they were slipping behind the Unix competition in terms of bang for the buck, and in fact I have pointed out many times that artificially high prices for iSeries and System i iron and software features like 5250 capacity enabled IBM to compete against Sun Microsystems (now part of Oracle) and Hewlett-Packard to drive down Unix prices for everyone and therefore give Big Blue the pole position in the Unix market. The trouble is, OS/400 and now IBM i shops could care less about Unix.
Look at the incremental cost of upgrading from one processor to two in the ProLiant machine. It is next to nothing. It only costs a couple grand to double the performance of the box moving from the G6 generation using two-core Xeons to the latest G7 generation using six-core Xeons. And there is no incremental base Windows or SQL Server licensing fee, either. You just pay for any additional users, if you have them.
What almost all OS/400 shops had for a decade and what almost all IBM i shops have had for the past five is a lot of Windows gear. They have experience with Windows, and even though they may not love it, I can assure you that they love their money more than they love the Power Systems platform running IBM i. And the only way to keep the IBM i platform going is to compete, year in and year out, and try to make it up in volume. Maybe IBM has a spreadsheet somewhere that says otherwise, that customers are stuck and a certain percentage will pay these high prices on the bigger Power 720s and the rest of the entry Power Systems line.
This is what happens when a company focuses on earnings per share and stock buyback instead of fighting to keep every customer. To whit, I say that Sam Palmisano and the rest of IBM’s top brass should be paid for every customer they keep and every new customer they gain, not on what the share price is doing down on Wall Street. If I wanted to be real mean–and as the editor of this newsletter, let me assure you, I most certainly do–I want Palmisano to be compensated solely on the number of IBM i customers he retains and get a bonus for the number he brings onto the platform. That ought to get the price lists where they need to be to compete.