The Lean, Mean RPG-5250-DB2/400 Machine (Continued)

<<< Click to return to the first part of this story

For the server configuration, I want IBM to get radical. First, let's talk about main memory. I think an 8 GB maximum should be enough, with a 2 GB minimum. I would scale memory up 1 GB for every 800 CPWs or so. That's simple.

Next, get rid of the Input/Output Processors (IOPs) as is the plan and let the processors and intelligent I/O Adapter (IOA) cards deal with I/O. IBM is planning on doing their with the future i5/OS V5R4 and is rumored to have a beta of this support in an interim release called i5/OS V5R3M5. I do not want any 5250 twinaxial controllers in this machine. You can use the 5250 protocol, but only through the dual, integrated Gigabit Ethernet ports in the system.

Now, for storage. I want an embedded RAID 1 disk controller on the motherboard if I can, or an outboard controller in a PCI slot if not. I want this server to use either Serial ATA (SATA) drives or small form-factor Serial Attached SCSI (SAS) drives. SATA drives are plenty good enough if they are mirrored, and there is not reason not to go with the smaller, cooler disks. IBM has been selling 2.5-inch "simple swap" (as opposed to hot swap) 80 GB, 160 GB, and 250 GB SATA disks in its xSeries servers and IntelliStation workstations. These SATA drives, which spin at 7200 RPM, cost $165, $219, and $329. The 2.5-inch SAS drives are a bit pricey, at $399 for 36 GB and $639 for 73 GB. They spin at 10K and are hot-swappable, though, so these might be interesting.

Still, I am interested in performance and price/performance here, so forget about the SAS drives. I want enough capacity and room to have a maximum of 1 TB of mirrored storage in the i5 System for SMB, so we need to use the SATA drives. IBM Rochester will just have to write some SATA drivers and get a SATA RAID 1 card. I don't want RAID 5, since it can cause some problems with i5/OS and its single-level storage if a RAID 5 disk group fails. RAID 5 adds a lot of cost--more than mirroring SATA drives. So forget RAID 5. The i5 System for SMB needs space for eight 3.5-inch SATA drives, then, which gives mirrored capacity. For a controller, try 3ware, which has a RAID 1/5 controller called the 9500S-8 that will fit the bill and which has FreeBSD and Linux drivers that can be hacked for i5/OS. 3ware also has hot-plug RAID drive cages, which hold four drives each (the RDC-400-SATA). That 3ware RAID controller costs around $500 in single-unit quantities. The drive cages cost about $200 each. So, for 1 TB of mirrored disk capacity, we are talking a street price of $2,632 for the drives, $500 for the RAID controller, and $400 for the cages, for a total of $3,532. To make configurations simple and because not every customer needs 1 TB of mirrored data, it might make sense to offer the 80 GB and 160 GB units as well. I don't think any configuration should sell with fewer than four disk arms, and mirroring should not be optional. i5 customers expect a a high degree of reliability. Now, keep in mind that IBM charges a lot more than this for iSeries controllers and disk capacity. Not on my i5 System for SMB line they don't.

For archiving and backup, which must be included in any i5 system, I think it is easiest to just use a single AIT-1, AIT-2, or AIT-3 outboard drive from a vendor such the GST, which supports iSeries and pSeries machines with its tape drive. For anywhere from $1,650 for an AIT-1 drive (91 GB max, 37 GB/hour speed) to $2,800 for an AIT-3 drive (260 GB max, 112 GB/hour), you can get a drive that will also work on servers besides the iSeries. Whatever drive IBM picks, customers only get one choice. Keep it simple, for reasons I will explain below.

The whole point of the i5 System for SMB is that it includes everything you need to run your core business applications. I don't want it to necessarily be an email server or a groupware server, although there are some reasons to do that. I am not trying to create a box that will be able to run a zillion partitions of Linux and AIX or use lots of IxS co-processors running Windows. Forget that. I want the i5 System for SMB to have three partitions: the primary i5/OS partition, the backup i5/OS partition (which provides failover capability if customers want to buy that), and a Linux partition to run a firewall or any other software that customers might need for infrastructure, such as the Apache Web server or various email and groupware programs (Scalix is pretty sweet, has native Outlook MAPI support, and costs a mere $60 per seat. I am seriously considering switching our company to it from SUSE Openexchange Server 4, which is a bit long in the tooth.) No Notes/Domino, no Workplace Services, no Exchange, no GroupWise. You want that stuff, buy the bigger i5 systems or buy Windows or NetWare or Linux servers. I think a 200 CPW to 400 CPW cap for the Linux partition should do the trick for most small businesses that have modest messaging and infrastructure requirements.

I want a few more things. First, I want a dashboard with a Web front end that does very simple things like: activate Linux partition or activate HA partition, or perform tape backup to archive data. Make this a turnkey solution, which means load it up with trial software from the OS/400 ISVs so I can just click a button and try any ServerProven software that has a trial version. Keep this list and the software behind it updated electronically so it is always current. I also want it to have automated PTF applications. Call it OS/400 Update for all I care. But get customers out of having to wade through patches. Finally, I want no extra charges for 5250 processing capacity. These machines provide full-on 5250 OLTP processing. Customers can use WebSphere or third-party application servers, WebFacing or Host Access Transformation Services or whatever, and these are also installed on the disk drives of the system and ready to be activated at the click of a button. Make it easy for customers to try and buy the wealth of software the OS/400 community has to offer.

Which brings me to pricing. I have been saying for some time that IBM needs entry machines that cost what competitive products cost. When you configure a current i5 520 Express machine, it is not difficult to get to a box with fairly modest performance (5,000 to 10,000 TPC-C online transactions per minute) for $20,000 to $40,000. This is roughly three times the price of a Wintel or Lintel server configured with a database and other tools--and it is also crazy. Only i5/OS Standard Edition machines offer anything comparable to Wintel and Lintel pricing--but they have no 5250 processing capacity. To get a machine in the same power class as my i5 System for SMB model 108, you would easily pay hundreds of thousands of dollars. Wintel and Lintel boxes--configured similarly with operating systems, databases, and like-for-like features--are about one-fifth to one-sixth the price. I complained about this last August, and I am complaining about it again this year. I think IBM is perfectly justified in charging a hefty Unix-sized premium for its big SMP i5 servers. But in the single- and dual-core server market, these prices just won't fly. IBM has to price more sanely. I want to cut that entry price in half, and put a lot more in the box.

One way to get around the pricing issue is to stop selling the boxes entirely. The i5 System for SMB would therefore not be something you acquire, but something you subscribe to.

If you could charge about $6.75 per CPW for a configured machine (which also assumes memory and disk capacity scales with the CPW performance in the configured machine), the list price to acquire these boxes would range from $10,800 for the i5 System 100 to $40,500 for the i5 System 108. IBM could charge a nominal fee of around $1,000 to activate a Linux partition and automatically install Red Hat Enterprise Linux 4 or Novell SUSE Linux Enterprise Server 9. (There's a 20 percent markup from list price among those Linux suppliers for their Power-based Linuxes, and IBM gets deeper discounts than that. That is enough profit.) Now hold your breath: This price includes i5/OS, DB2/400, RPG, and COBOL. There is no price of performance penalty for using the 5250 protocol, although you do not have to use it. You have to pay extra for WebSphere or HATS or WebFacing, or the third-party Web-enablement tools. Activating the HA partition should be free, but you have to pay for special editions of the HA software from DataMirror, iTera, Lakeview Technology, Maximum Availability, Trader's, or Vision Solutions. The exact nature of the HA such small companies need is unclear, but simple failover inside a single system is probably better than nothing at all.

Monthly rental fees for these machines would range from $300 to $1,125 for the base configurations, and customers would have to commit to a three-year contract with IBM to get one. This contract absolutely has no penalties for upgrading. In fact, it will assume and encourage this, because the goal of these feeder systems is to get companies on vintage AS/400 and iSeries servers to modernize their hardware and software and then, when they get more sophisticated, move to the real i5 line. IBM has to set monthly rental prices for all of the components that go into an i5 System for SMB, and any upgrade is a net-cost upgrade. That means no mark up if customers move from, say, a 2 GB to a 3 GB memory block or from 1,600 CPWs to 2,400 CPWs. And to make things simple for IBM, list price is the street price. No discounting whatsoever, not even once. Buy one, get one; buy ten, get ten. These prices are so low, no one should complain.

The only stickler to this plan is that IBM will have to sell about four times as many boxes to make the same money as it does on much more pricey iron today. I would make the argument that IBM is not actually selling machines to SMB customers at all, but rather only to existing Ms in the SMB space who already know what an iSeries is. To my way of thinking, this is net new business--and business the iSeries Division should have been chasing for years.

Next week, I have another idea on how to take these i5 Systems for SMB and make a more expansive i5 product line--one that goes right to the heart of the marketing that Wintel and Lintel server makers and Oracle have been spouting for the past several years. I want to take the fight to their turf and fight an offensive maneuver, not a defensive one.