The Lean, Mean RPG-5250-DB2/400 Machine

by Timothy Prickett Morgan

Microsoft has been getting a lot of attention for its future consolidated line of application software, dubbed Dynamics and created under the code-name "Centro." Dynamics will be launched on top of its "Longhorn" Windows Vista platform starting next year, and it will include the future Exchange Server 12 release of the company's groupware and messaging products as well as accounting software. Dynamics will be tightly integrated with Microsoft's Office suite, and by 2008 will merge the code bases of its Great Plains, Axtapa, Navision, and Solomon product lines. This is a serious move by Microsoft, and one that the iSeries has to counter with more than rhetoric.

Rhetoric is, unfortunately, all that I have as a means to foment change, and your voices are the only things that will compel Big Blue to take action and make difficult short-term choices that will benefit the OS/400 platform over the long term. By doing so, IBM would help all of us--itself, the OS/400 community, and the companies that use the OS/400 platform. I have a few ideas about how to beat Microsoft, and because Big Bill is not moving all that fast on the Dynamics rollout--Longhorn Server is not expected until 2007, after all--there is a chance--a very slim chance--that if IBM acts now, it can knock the wind out of the whole Dynamics initiative, or at the very least blunt the damaging effects it will have on the iSeries business. And make no mistake, the Midrange Alliance is child's play compared to what Microsoft will do with Dynamics. Having failed in acquiring SAP about two years ago and missed its chance to acquire the former PeopleSoft, Microsoft wants to become the dominant supplier of application software for both the S and M in the SMB space. (Hey, get your mind out of the gutter. I know it is Rome Week on the History Channel and that seedy Rome series is running on HBO, but this is a family newsletter.)

And as I have said so many times before, IBM has to get back to the fundamental, basic, intrinsic idea of the Application System/400 platform: providing affordable ERP software for small and medium businesses. This is obvious, right? And notice how I said affordable. What is not so obvious is that maybe IBM should stop telling OS/400 shops all about Linux and Unix and Java and now .NET and how to modernize their applications and start telling the over 200,000 customers who use OS/400 to support their RPG and sometimes COBOL applications and the 5250 protocol that it is OK to like and use OS/400, DB2/400, RPG, COBOL, and the 5250 protocol. Come on, Mark Shearer and Peter Bingaman, say it loud and say it proud: OS/400, DB2/400, RPG, COBOL, and the 5250 protocol augmented by WebFacing and similar third-party tools are perfectly acceptable for building applications. Clap your hands together as you say it with me, brothers, and keep on saying it until you actually believe it. Come on, Frank Soltis, you been waiting to sing along, too. Jump right in on the next refrain. And Bill Zeitler, I wanna hear you sing it so Sam Palmisano can understand the words. We got the thousands of voices of the Rochester Labs and Tabernacle Choir backing us up, filling in the sharps and flats, so don't be shy.

As you are singing the praises of the once and future AS/400, I want you to look at your iSeries pricing. Your pricing doesn't sing the praises of what I would call the native OS/400 technologies, now does it? It says great things about consolidating Linux and AIX workloads on the iSeries and getting Windows on Integrated xSeries Servers and having OS/400 play nanny to Windows (and boy does Windows need something to keep an eye on it). I know server consolidation is important to IBM in general and the iSeries marketing plan in particular, but I think you would have a much easier sell to vintage AS/400 and iSeries shops if you would start giving them what they want--cheap iron, the assurance from IBM that RPG, DB2/400, and the 5250 protocol are viable alternatives to Java and .NET, and pricing that not only stops penalizing customers who use the native interactive technologies of the AS/400, but maybe even rewards them for supporting this technology. This technology may be old, but it is arguably a lot more efficient than .NET or Java.

With the Power5+ processors coming out in the spring, probably some time around March, there is still time left to do something radical, something crazy, something totally aimed at the heart of the Microsoft-Intel duopoly. I want IBM to do something that will allow companies with homegrown code to defend against the idea of using Microsoft Dynamics on Wintel iron, to give them the confidence that they can get a cheap iSeries box and keep their code and their jobs maintaining it because, after all, this is the heritage of a lot of OS/400 shops. It is OK to write your own code, and it is OK to do it in RPG and to use Web-enablement technologies to make it all pretty. (You don't think .NET and Visual Basic or Java and J2EE are any better, in a technical sense, than 5250 and WebFacing or HATS or WebSmart or RPG Server Pages or RPG Smart Pages--just to name a few of the possibilities--do you?) I want IBM to do something that also provides a real entry iSeries system that is an appropriate platform on which third-party application developers can create very inexpensive and high-performance OS/400-based applications and, better still, complete turnkey solutions. Yes, I know how crazy that must sound. But hear me out.

Unlike Microsoft with Windows, IBM owns the whole OS/400 platform: the iron, the operating system, the database, the middleware, and the compilers. Microsoft doesn't own the iron in the Wintel stack, and that gives it a disadvantage. IBM can price the i5 hardware or software any way it wants to, and what I want IBM to do is to stop selling the box and start selling a service.

The System i5 for SMB

To simplify things, I want to give this new i5 box a name. And in keeping with its latest mainframe announcements, I think the System i5 for SMB is a good name for this inexpensive, integrated, turnkey box that unashamedly uses native AS/400 technologies. No i5 500 Express Plus Minus Anything or This And That. I don't want to talk about WebSphere Express Gizmo Plus This That and The Other, or any other silly name that doesn't stick in anyone's head. IBM has gone absolutely nuts in its naming schemes, as has Sun Microsystems when it puts Java in front of everything these days, whether or not it is based on Java or has anything to do with Java. This talk is nonsense--even if the technologies are great--and it gets in the way of selling a system to people who don't want to know how the car works, they just want to drive it.

So with the System i5 for SMB, let's start with the processor. With the Power5+ chips coming out in the spring, IBM is moving from 130 nanometer copper/SOI chip technology to 90 nanometer copper/SOI/strained silicon technology. If IBM uses that shrink to let the Power5+ chip consume less power and generate less heat--rather than to crank up the clock speed of the chip--then you can package a Power5 chip into a slightly smaller package. Say something that looks a lot like an xSeries tower server outfitted to hold a single-core Pentium 4 or dual-core Pentium D chip from Intel. The current Power5 chips used in the iSeries line run at 1.5 GHz and 1.65 GHz and deliver about 2,400 CPWs (no L3 cache) and 3,300 CPWs (with L3 cache), respectively. If IBM used the 1.9 GHz Power5s, it might have only hit 3,800 to 4,000 CPWs and generated a lot more heat, which is why you didn't see the iSeries opt for that chip. The yields are also lower on the higher clock speeds, and IBM needs the fastest chips for the pSeries line so it can compete against other big Unix iron. I say, let the Unix guys have the fastest Power5 and Power5+ chips, and to hell with speed. In fact, with the Power5+ chips, I want IBM to use the slowest chips that will still function in a new i5 tower. I don't know what that speed is, but it is probably between 1 GHz and 1.5 GHz. Say IBM can deliver these two clock speeds easily with the Power5+ chips, and say further that chips running this slowly might not have even made it into products because they aren't impressive enough. I am not trying to impress anyone except a customer who wants a cheap i5 machine that is a better deal than a Wintel configuration.

Assume further that we can use the dud chips where only one core is working--these might have also ended up in the scrap heap. So we have four possible configurations then, and IBM offers simple and net-cost processor upgrades between them: a single 1 GHz core, a single 1.5 GHz core, two 1 GHz cores, and two 1.5 GHz cores. If the L3 caches add to the cost, kill them, because we are not going to care much about Java performance on this box. With no L3 caches, that would be a performance spread of 1,600 CPWs, 2,400 CPWs, 3,000 CPWs, and 4,300 CPWs, and with the L3 caches it would probably be a spread of 2,000 CPWs, 3,000 CPWs, 3,800 CPWs, and 5,400 CPWs. The L3 cache helps a lot on the two-core configurations, not so much on the single-core versions. To get nice granularity, you could offer all seven CPW performance levels and all eight configurations, of course. I'll do this just for the sake of argument. And just to make a big top-end machine and to make the naming convention work out, let's add in a version based on a two-core box running at 1.65 GHz with the L3 caches on. Here's what such a lineup might look like:

To put this into perspective, this product line would span the performance of tower X64 servers using 64-bit Pentium 4 chips from Intel or Opteron 100 Series chips from Advanced Micro Devices up through the current crop of single-socket, dual-core Pentium Ds and Opteron 100s and continuing up to the single-core Intel Xeon DPs (used in two-socket machines) and the Opteron 200 Series (used in two-socket servers as well). These are the absolute sweet spots in the X86/X64 server market: entry single-socket servers and entry two-socket or two-core servers. This is where Dell and Hewlett-Packard get most of their sales volumes and a somewhat smaller share of their server sales. Aim right for it. Don't go for a glancing blow.

Click to continue reading the next page of this story >>>