IBM’s Battle Plan for i5/OS Blade Servers
February 18, 2008 Timothy Prickett Morgan
Putting together a blade server strategy that will appeal to the i5/OS and OS/400 ecosystem is not as simple as getting a blade server into the field and making OS/400 and i5/OS run on a Power blade. If it were–and I will remind everyone that this is how Windows and Linux got blades, and eventually Solaris, HP-UX, and AIX, too–then you would expect that OS/400 would have been on a blade server many years ago and the blade form factor would account for roughly 10 percent of revenues among i5/OS shops today, as it does for the server market at large.
If IBM had started early with blades for OS/400, then forward-thinking OS/400 shops, who have never been afraid of paying a premium for advanced technologies that simplify their IT lives, might have pushed the penetration of blade form factors as part of overall iSeries and System i sales higher than the ratio of blades to all other servers in the industry at large.
Before I get into IBM’s marketing plans to peddle the new JS22 Power6 blade server to i5/OS and OS/400 shops, I want to go over a few items that are on my mind first.
To begin with, I have always been somewhat baffled about how IBM even made the decision to remove the PowerPC AS instructions from the PowerPC 970 family of chips to begin with, and had it not done this, then IBM would have been selling single-, dual-, and quad-core Power blades in the BladeCenter running OS/400 and i5/OS for as many years as it has been selling AIX blades. That would be starting with the JS20 blade from November 2003 (using single-core PowerPC 970s), rolling up through the JS21 blades in February 2006 (using dual-core PowerPC 970MPs), to the JS22 in November 2007 (using the dual-core Power6 processor). I would have thought that if IBM could not get a PowerPC blade that supported the PowerPC AS instructions and memory tags necessary for running OS/400 and i5/OS into the field, it would do the smart thing and gear down some Power5 chips so they could do the job and not overheat the box. Even an old 600 MHz S-Star or 750 MHz I-Star PowerPC chip, which definitely supports OS/400 and i5/OS, would have been preferable to nothing at all.
While I applaud IBM for finally getting a blade into the field that can run i5/OS V6R1, this took far too long to come to market. In fact, it won’t really be here until V6R1 ships on March 21. Moreover, the four-core Power6 blade, which is in the P20 software tier, is a lot more computing power than many customers running i5/OS and OS/400 workloads need. IBM has yet to provide performance information on the JS22 blade server, but a Power6-based System i 570 server running 4.7 GHz chips, which have 4 MB of L2 cache per core and 32 MB of L3 cache for each core pair, delivers from 5,500 and 21,200 CPWs of i5/OS workload performance. Dropping down to 4 GHz clock speeds on the four cores in the JS22 blade and removing the L3 cache memory (which IBM almost certainly had to do to keep it from overheating the blade) probably puts the performance of the J222 blade with a single core activated at around 3,800 CPWs and a four core configuration at somewhere around 14,500 CPWs. Considering that the four-core 570 has access to 192 GB of main memory, compared to the 32 GB of the JS22 blade, performance could be significantly lower for memory-hungry applications. (CPW benchmark results are usually based on maxxed out configurations, and the JS22 has to be memory constrained–again, for heat reasons more than anything else.)
Customers only have to activate one of those cores for running i5/OS, to be sure, but that is still a lot more oomph than SMB customers–the kinds that use deskside AS/400, iSeries, and System i servers, need. I have heard talk of a single-socket JS12 blade server, also based on Power6 chips, and I think this is the blade IBM needs for the i5/OS market. And I think that for many good reasons, not the least of which is that the P20 software tier–which is justified based on the System i 570 using Power6 chips, which is in the P30 tier with four cores–is nonetheless out of reach for a lot of small shops who have applications in the P05 or P10 tiers. What IBM needs is a JS12 blade with one core activated that is priced in the P05 tier and like the 515 user-priced server and a JS12 blade with two cores activated that is in the P10 tier and is priced like the 525 machine.
I know that IBM is positioning this blade for massive consolidation, but it has to go smaller to meet where the customers are. Take a look at this feedback I got from one iSeries shop in the wake of the JS22 and V6R1 announcement:
“I just wanted to pass you a note on our shop. We have one iSeries 810, two Linux servers, and one Windows server. We were drooling over the blades with i5. But if you are correct and I assume you are, IBM pricing blew it. I’m not really worried about that first $3,000 bump on the blade. Now way in h*** I’m paying P20 prices! Not going to happen. Our lease is up the end of April I’ll negotiate a bargain on the 810 and keep it. S*** IBM.”
Another satisfied customer, eh? The $3,000 bump this reader is referring to was the premium IBM was charging for the i5/OS blade compared to the AIX blade, and it was an error on the IBM Web store, which I told you had been corrected last week internally at IBM and now is actually correct on the Web store. The premium is still $1,307, but that might have more to do with the SAN connection features. I can’t be sure, since IBM doesn’t actually spec things out on the Web store. (Which means IBM is not really selling it online, right?)
Another thing that I am annoyed by is the fact that i5/OS is not yet mentioned as a supported operating system on the BladeCenter blade servers on IBM’s site, which you can see here. This page doesn’t show i5/OS V6R1 as a possible option for the BladeCenter blade servers. It does mention AIX, of course, and Linux, which IBM loves, as well as Solaris from Sun Microsystems and Windows from Microsoft. Now, I know how hard it is to keep a Web site updated–believe me–but IT Jungle has five employees compared to IBM’s 355,800 employees. I think IBM has more resources to bring to bear and can do a better job promoting its own i5/OS platform on its own site.
Now that I have got that off my chest, let’s take a look at what IBM is telling its business partners about how to sell the JS22 blades running i5/OS.
First, some statistics. According to IBM’s own internal research, some 80 percent of shops have one or two OS/400 or i5/OS systems; this means that 20 percent have more than three machines in their shop. Moreover, IBM believes that 85 percent of the machines installed at OS/400 and i5/OS shops have only a single processor core in them; this again implies that only 15 percent have multicore boxes. Now, here’s where it gets interesting. Among the same i5/OS and OS/400 shops, the average number of Windows boxes is 12 boxes, and two-thirds of them are not bearing the IBM label. That is a lot of Windows iron, and IBM says that most of these machines are running infrastructure workloads–Web, print, file, and email serving–and are wickedly underutilized. Each server tends to be a standalone silo, and none of the storage on these i5/OS, OS/400, and Windows boxes have been consolidated onto storage area networks.
Now, rather than take on all of those Windows boxes and try to move them to a System i box with Integrated xSeries Servers, which is the tactic IBM has been using since the File Serving IOP that dates from, what, 1993, IBM plans on working these i5/OS and OS/400 shops from the Windows angle with a consolidation play based on putting VMware‘s ESX Server hypervisor on a few X64 blades and moving those dozen average Windows servers onto a few blades. ESX Server requires a shared storage array, which helps IBM’s disk array sales, and going virtual on Windows makes the setup more resilient and more efficient. According to IBM’s presentation, the positioning of “i5/OS on Blade is incremental,” adding value to the overall solution and giving IBM a competitive differentiator. But that is not the major push. IBM’s real goal, according to the presentation, is to turn what would have been a $25,000 upgrade opportunity for a System i box into a $100,000 BladeCenter transaction that includes some System i action.
IBM provided partners with a few scenarios to think about as they approach JS22 sales. In one simulated customer scenario, the shop has two AS/400 270s, 14 vintage Hewlett-Packard rack and tower servers, and no storage area network; IBM suggests that this customer buy a BladeCenter H chassis, a JS22 blade for running i5/OS V6R1, from five to 10 HS21 Intel-based blades, and a DS4700 storage array to provide SAN capability for the whole shebang. This seems reasonable. But in a similar scenario where the customer has two AS/400 720s and has 10 new HP servers and already has a SAN for the Windows boxes, IBM is suggesting merely that the customer move the OS/400 workloads up to i5/OS V6R1 and newer iron. At a more modest shop with a single AS/400 270, three old HP servers, and no SAN, IBM isn’t going to try to peddle the JS22 anyway. The recommendation there is to get a System i 515 running i5/OS and use iSCSI links to newer System x servers. The BladeCenter H is overkill for these shops, so is the JS22, and IBM knows it.
Of course, a BladeCenter S chassis with a SAN module, a single JS12 blade, and two virtualized X64 blades would do the trick just fine–except that the JS12 blade doesn’t exist, and if it did, it does not support the BladeCenter S small chassis.
I remain hopeful that the top brass at IBM will think this all over and tweak the product line and strategy a bit to get the right products out the door. IBM did a good thing with the System i 515 and 525 and the repackaging and repricing of hardware and software on the System i 570 last year. Hope is cheap, even if the blade iron still isn’t.