The Four Hundred

In last week's issue, I told you that IBM had, as part of the OS/400 V5R2 and iSeries Model 890 announcements, significantly broadened and improved its Capacity Upgrade on Demand offering for the iSeries. I then told you that this week I would give you an advanced look at some of the CUoD technologies that IBM is cooking up for other aspects of the iSeries beyond processors. IBM's got some pretty cool things that it might do.

Before we get into that, let's review what IBM is already doing. As I explained last week, with capacity on demand, an IT vendor ships extra processors or storage in a server, and these resources can be activated with special keys provided over the Internet by the vendor when increasing workloads demand it. Customers do not have to take their machines offline, install extra processors or disks, and then reboot their servers; they just turn the key (not literally, but metaphorically), the vendor kicks out an invoice, and they pay for the capacity they have added. This is much easier than doing an upgrade. In addition to capacity on demand, IBM offers selected customers capacity standby, which means that additional servers and storage are kept on site in the event that these machines are needed. A vendor cannot really do capacity on demand with uniprocessor, for instance, since there are no extra CPUs to activate when demand spikes. So, instead, most big server vendors offer customers the option of paying a slight premium to keep spare servers and storage on site.

Since OS/400 V4R5, IBM has offered some capacity on demand features in the iSeries Model 840 server, and customers opting for this method of acquiring those machines paid a 15 percent premium for processing capacity, compared with buying a server and doing traditional incremental upgrades. IBM is now offering its CUoD packaging and pricing on all Model 830, Model 840, and Model 890 servers with four or more processors--and there is no premium for these CUoD features. All of the machines in this group of iSeries models are CUoD-enabled, but IBM is offering customers the choice of ordering machines with CUoD preconfigured from the get-go, to speed up the capacity on demand process and to track the popularity of this option. Customers can still buy non-CUoD iSeries machines, but why would anyone do that? It's silly.

Under the current CUoD scheme, customers can activate some or all of the spare capacity in these machines for 14 days on a trial basis. No IPL is required in order to do this. The CUoD clock keeps track of the time when the CUoD capacity is activated and automatically decommissions it after the 14 days are up. Once a customer buys additional CUoD capacity in the iSeries server, the trial settings are reset and the machine is then available for another 14 day trial period in the future. Once you decide to activate all or a portion of this excess capacity, IBM posts a CUoD key unique to your server on a special Web site and also mails it to you.

According to sources familiar with IBM's thinking on CUoD for the iSeries, Big Blue is examining yet another option for capacity on demand. One of the drawbacks of capacity on demand from all IT suppliers is that it is a one-way deal: You can activate and buy capacity, but you cannot deactivate capacity. In a more forgiving world, it would be possible to temporarily rent spare processing capacity when, for instance, you need to run an old and a new set of applications side-by-side during the course of an upgrade. IBM is reportedly evaluating just such a temporary CUoD activation option for the iSeries, which is a good idea. The way it will work, apparently, is that it will allow customers to temporarily activate capacity for a set number of days for just the kind of scenario outlined above. The pricing for that capacity will, I am guessing, look very much like a leasing agreement for a whole iSeries machine, with the terms being over the course of days, rather than years. As long as IBM keeps the activation and accounting systems behind this temporary CUoD offering simple, this could actually drive up iSeries processor sales. The "potato-chip effect"--you can't have just one--will prevail, and as long as the spare capacity is inexpensive, people will use it. This is a sly, as well as a helpful, means of providing upgrades, and one that will probably become the norm over the long haul among all server vendors. The server business will, in effect, come full circle to the 1960s and transform back into a rental business.

IBM is also reportedly looking into expanding CUoD from covering processing capacity on the iSeries to also include units of memory and disk on demand. This makes perfect sense, since a balanced machine requires that new processors have a certain amount of memory and disk capacity associated. For many workloads, if you just add processors, all you do is shift the bottleneck from the processor to the memory subsystem or disk subsystem, and not a lot more work gets done.

The system designers in the Rochester iSeries development labs are also said to be examining the possibility of selling units of 5250 interactive processing capacity on demand, which I think is a good idea, too. This would, presumably, offer finer granularity than the existing 5250 interactive capacity brackets that are provided through the existing set of auxiliary interactive feature cards, which tell the OS/400 CFINT governor how much of an iSeries' processing capacity to dedicate to interactive jobs. I would not be surprised if IBM eventually switched from governing 5250 interactive capacity to governing database and OS/400 IFS access, and, if the company does, there will probably be CUoD for these aspects of the iSeries server as well.

Exactly when or if these potential CUoD offerings will see the light of day is unclear. All IBM is saying is that these options are being considered for a future release of iSeries hardware and OS/400 software.