i5 Model 595: Big Bang for Big Bucks
November 8, 2004 Timothy Prickett Morgan
Most server lines have a bell curve that describes their price/performance. Because of intense competition, entry servers offer the best bang for the buck. And because big customers that buy big servers stamp their feet and get coddled, high-end servers tend to be priced very aggressively, too. It’s the customers in the middle, which represent the vast majority of the installed base for OS/400 and Unix servers, that tend to get the worst deal.
IBM‘s pricing for its biggest AS/400 and iSeries boxes was very aggressive, particularly since these machines were put through the paces of the TPC-C online transaction processing benchmark to prove the value of the OS/400 architecture. In fact, in 1992, the AS/400 E series machines were the first to run on the TPC-C test, and they demonstrated the great value of the OS/400 platform without bringing into account the total-cost-of-ownership and ease-of-use issues that have been central to marketing in more recent times. Back then, AS/400s also had another feature, which I will call “ease of marketing.” The AS/400 offered reasonable bang for the buck, and you got architectural benefits to boot. Around the time of the CISC-to-RISC transition, in early 1996, the AS/400 stopped keeping pace in the server price wars, even though the AS/400 often showed the best performance IBM could muster on any server platform, including its beloved (and I think often overrated) mainframes. By the time of iSeries generation, the OS/400 platform had fallen so far behind (even after having 5250 costs extracted in server editions and then the OS/400 Standard Edition models) that IBM stopped running TPC-C tests on the top-end systems. And if you think this doesn’t relate to eServer i5 Model 595 pricing, you are wrong.
IBM’s last TPC-C test on a top-end OS/400 server was for an iSeries Model 840-2420, which was a 24-way box based on the 500 MHz I-Star PowerPC processors. This server was rated at 16,500 CPWs and delivered 152,346 transactions per minute (TPM) on the TPC-C test at a cost of $44.52 per TPM. After price changes on the iSeries line for processors and memory in 2002, IBM could probably deliver this machine for about $35 per TPM. That was about twice the cost of a competing Unix server.
You see, IBM may have merged the OS/400 and AIX server lines, but it has bowed the OS/400 server out of the intense competition at the high end of the server market, just like it has done with its zSeries mainframes. And Model 595 prices reflect this lack of competition. IBM is betting that it can charge big bucks to the Model 595 for the relatively few customers who will want it, and its high list prices for the machine seem to be set so that IBM can offer big discounts and still charge a lot of money for this machine. That said, as the biggest i5 machine, a fully configured Model 595 still gives much better price/performance on 5250 workloads than lighter configurations of the Model 570 or Model 595. The problem is, I can only think of a dozen customers who need 165,000 CPWs of green-screen processing power, and these customers would have to consolidate all of their AS/400 and iSeries iron onto this big box. Considering that the 64-way Model 595 is in the P60 software tier and partition-based pricing for application software is not yet the norm, the concept of using the Model 595 as a giant server consolidation box seems to be a bit ahead of the practical economics of it.
Now it may sound like I think the Model 595 is a bad product to bring to market. I don’t. I just think that OS/400 software tiers are looking a bit 1980s right now, and that this is a problem. And I think that pricing for the Model 595 is pretty decent compared to earlier iSeries machines if (and this is a big if, I know) customers activate all 64 cores in the biggest “Squadron” Power4 box and turn on OS/400 and 5250 Enterprise Enablement features for all of those cores. And that is why, in the iSeries and i5 big box salient characteristics table that accompanies this story, I have configured an eServer i5 Model 595 Big Bad Box. This machine, which has 10 times the processing capacity of the last machine that IBM tested on the TPC-C test, almost four years ago, offers better value on green-screen applications than the prior high-end iSeries Model 890, and even better than the new i5 Model 570. The bang for the buck of that top-end machine is what I think customers should get across the entire i5 line, in fact. It is high time to flatten that price/performance bell curve and give the core OS/400 midrange customer as good a deal as IBM is willing to give to the biggest OS/400 shops.
In the table I made, I have created two different configurations. One compares the iSeries Model 890, the i5 Model 570, and i5 Model 595 with green-screen processing capacities in the range of 25,000 CPWs; next to these is the i5 Big Bad Box. In the other comparison, I activated more processors, OS/400 licenses, and 5250 features in these three machines and pushed the aggregate green-screen processing power up to 35,000 CPWs. Again, the i5 Big Bad Box is there for the purpose of reckoning what a great box might look like. Take 25 percent or maybe even 35 percent off the list price of a heavily configured Model 595 shown in the table, and you probably are close to the street price.
In the 25,000 CPW scenario, the Model 595 with right Power5 cores running at 1.65 GHz is actually 14 percent more expensive than a similarly powered Model 890 with 20 1.3-GHz Power4 cores. I have configured these machines with 64 GB of main memory and OS/400 i5/OS and 5250 enablement features on each box. (In comparisons IBM has been using to show the improvement in price/performance of the i5s versus the iSeries, IBM has been comparing the iSeries machines with full 5250 activation against Model 570 and 595 boxes with only four cores or 12,000 CPWs of green screen processing turned on. IBM is, however, turning on OS/400 for the extra cores in the box, but not the 5250 support. This is not, strictly speaking, a fair comparison, even if it is true that very few customers need more than 12,000 CPWs.) In that 25,000 CPW scenario, the Model 570 is clearly a better deal than the Model 595, but the i5 Big Bad Box offers 42 percent price/performance improvement over the Model 890. This is what happens when IBM charges a flat fee for 5250 processing beyond seven processors, which it has done on the Model 570 and 595 servers.
In the 35,000 CPW scenario, a Model 595 server offers only about 8 percent better bang for the buck on green-screen workloads than the Model 890 it replaces. A 12-way Model 570 offers 16 percent better value, mainly because the base 570 box costs a lot less and so does its memory and Software Maintenance fees.
I think the competitive position of the OS/400 platform would be enhanced significantly if IBM got back into the TPC-C game and priced the top-end i5 servers to compete against big Unix and Windows boxes. If there is one thing I want for the new year, it’s a radical change in OS/400 server marketing. And I do not mean TV ads. I want IBM Rochester to be allowed to get back in the game. I want it to compete, and win, just like it used to do. If the i5 gets back on the server price/performance curve, then maybe, just maybe, sales will improve after IBM demonstrates to the world through action–not words–that the company intends for the OS/400 platform to compete aggressively in the midrange, not just turn it into a small mainframe with its exotic pricing that bears no relationship to Windows, Unix, and Linux platforms. If there is any reason why the iSeries is a push, not a pull, product, it’s that IBM’s pricing is out of step with the market, and everyone knows it. All of the total-cost-of-ownership and ease-of-use arguments in the world won’t convince more than the OS/400 faithful that the machine is worth the money, even if it is true.