The Four Hundred

I've got lots of feedback on the iDeal iSeries, and I will share as much of it as makes sense once I am done with the story series next week. In past iDeal iSeries stories, I've created the hypothetical hardware and suggested prices for them as well as for memory, disk, disk controller, operating system, database, and green screen connectivity components. This week, I'll bring it all together to see how these machines would compare to the real iSeries machines from IBM.

Last week, I crabbed about what IBM charges for disk and memory prices on the iSeries and AS/400 lines, but I didn't say what prices I would set for these features. My assumption is that the low-end iDeal iSeries machines, which I am basing on the future PowerPC 970 processor, would use common memory DIMMs and controllers used in entry Intel, Apple, and Unix servers, and therefore IBM can charge very low prices for them. On the iDeal iSeries machines that are equivalent to the current Power4 and Power4+ pSeries Regatta server lines, let's just set memory prices to the same level as on the pSeries and see what happens. I think setting disk and controller prices on the iDeal iSeries machines based on the PowerPC 970 equal to what IBM charges on the xSeries 440 is fair, and setting disk and controller prices on the iDeal iSeries machines that are identical to the current pSeries at the same price as those pSeries machines is consistent even if it is not, strictly speaking, fair.

When I put together base configurations and reasonable maximum configurations, I ignored feature cabinets, tape backups, expansion chassis, cables, and other features that are necessary for comparing real servers. Let's just see what the core central electronics complexes would cost, and then add users and software to the boxes and see how these compare to the real iSeries machines.

The prices for base iDeal iSeries configurations are very similar to the base list prices I showed two weeks ago, since my base configurations have the minimum practical memory and disk capacity on the machines to load the operating system. (Adding hardware to support users is something a true comparison would require, but doing so distorts the comparisons across processors because memory and disk are such a large component of server pricing these days.) I've added RAID disk controllers and disk drives to the iDeal iSeries machines to give us a starting comparison point. Based on the prices outlined above, you can add memory, disk, and disk controllers (eight drives per controller is a good idea, but they can do double that if you want to conserve money) to these base machines to beef them up.

Coming up with the practical estimates for base and maximum users for each machine is not a simple thing, but it is possible to come up with an estimate that makes sense. IBM's CPW relative performance rating is based on the TPC-C online transaction processing (OLTP) test. If you take the CPW ratings and multiply by 9 you get about what IBM will post on the TPC-C test for any AS/400 running V5 software. This is, however, for a three-tier configuration that offloads application servers to auxiliary machines. The machine given the TPC-C rating in a real TPC-C benchmark test is actually only running the database side of the TPC-C application. This is not how real OS/400 shops use their AS/400 and iSeries servers. The application portion of the TPC-C test accounts for about 20 percent to 25 percent of the overall workload if you go to a two-tier configuration (that's with the database and application servers as one OS/400 tier and an intelligent client as the second tier). So, if you take CPW and multiply by 9 and then by 0.8, you get the maximum number of TPC-C users in a two-tier configuration that can be supported on a given machine. (This is a rough equation that fits within a few percent of any TPC-C test run in the past few years.) Now, TPC-C is a peak performance benchmark with end users working constantly on a mix of relatively light transactions that are hitting against a database that is largely stored in memory. This is not how real end users work and it is not how real databases are accessed by real OS/400 applications. This is why I didn't base the number of users in my user-based software pricing calculations on the TPC-C test.

I used experience instead. I reckon that a more accurate indicator of the practical number of base users on a given machine is to take the CPW rating for that machine and divide it by a factor of 20. To get maximum users, divide CPW by a factor of 5. This fits roughly with the empirical data I have from talking to OS/400 shops over the years. (If you have a better way of reckoning users, don't hesitate to tell me. I welcome any suggestions. I will use the same scaling factors for both the actual iSeries and iDeal iSeries machines, so the point I am making with my arguments won't change.)

To make comparisons within the iDeal iSeries line and to a few select models in the real iSeries product line, I have created two different iDeal iSeries configurations: a reasonable base configuration that a customer would likely buy, and a maximum configuration suitable for a relatively heavy number of OLTP users. I can't make any warranty that the maximum configurations would support the maximum number of users, but I am at least willing to take a stab at the problem to show how budgets would be affected by the pricing model I have created. This is not a perfect scenario--regardless of the name I have given these idealized OS/400 servers--but it is a starting point for illustrating an alternative way of doing things.

To see the table of iDeal iSeries base machines, including their hardware, OS/400, database, and connectivity costs, click here. Note: when I say base configuration, I mean the smallest practical configuration IBM should sell for OLTP workloads, not a bare bones machine. This base machine does not include the cost of expansion chassis, tape backup, connectivity features, and so forth.

To see the table of iDeal iSeries maximum configurations, including their hardware, OS/400, database, and connectivity costs, click here. Note: when I say maximum, I mean maximum practical configurations for OLTP workloads, not a fully loaded machine.

Here's the deal. The base iDeal iSeries machines will, by my reckoning, support from 30 to just under 2,000 users right out of the box. Base hardware prices range from a low of $5,837 for a uniprocessor iDeal iSeries with a single 1 GHz PowerPC 970 processor, 1 GB of main memory and 72 GB of disk (capable of supporting about 30 users) to just over $2 million for a machine configured with 32 1.3 GHz Power4 processors, 128 GB of main memory, and 180 GB of disk. (Yes, I know that this machine is woefully light on disk, but I wanted to isolate processor and base memory costs.) When you do the math, this works out to between $150 to $200 per user for iDeal iSeries hardware at the low end for machines using the PowerPC 970 chip. With hardware pricing on the iDeal iSeries Model 831, 851, 871, and 891 machines set to the same prices as the pSeries Regatta equivalents from IBM, base hardware costs for base configurations range from about $250 to $300 per user on the Model 831 and Model 851 (equivalent to the existing pSeries 630 and 650 machines), to between $600 and $750 per user on the Model 871 (same as the pSeries 670), to between $1,000 to $1,125 on the Model 891 (same as the pSeries 690).

To price out the software, I put the standard, advanced, and enterprise editions of OS/400 and DB2/400 on different machines. The PowerPC 970-based machines got the least expensive OS/400 and DB2/400, the standard edition. The Power4-based machines that scaled only up to 4 or 8 processors got the advanced edition, and the bigger Power4-based machines that scaled up to 16 or 32 processors got the enterprise editions. (A reminder: You can see a table from last week's issue that shows how IBM currently prices OS/400 and DB2 components and how I would set prices by clicking here.) Simply put, I think a flat fee for OS/400 based on the number of CPUs and the OS/400 edition is sufficient; for DB2, there's a CPU component and then a user component based on either ODBC/JDBC access to the database or 5250 access to the database. The latter costs $225 per user, while the former costs only $50 per user.

When all the calculations are done in my pricing scheme, it costs from $255 to $1,238 per user for base iDeal iSeries machines that have only ODBC/JDBC access to their databases, and from $430 to $1,413 per user for end users who only access the base machines from the 5250 protocol. The hardware pricing is, with such relatively few numbers of users on these machines, the biggest factor in determining user pricing. The point in making these massive tables is that it shows the practical entry price points for customers of all sizes--small, medium, or large--into the iDeal iSeries base. No total cost of ownership argument here. With the iDeal iSeries, lower TCO is an attitude and a gift. All we want people to do is buy the machine and love it, rather than buy another machine and hate it.

This pricing scheme is intentionally progressive--those customers who use more hardware and software features and who require IBM to invest lots of research and development into big servers and advanced database technologies end up paying more money than those who do not. It's like that in the real world, and iSeries pricing should reflect this.

The other thing you will note is that this iDeal iSeries pricing model keeps the cost of adding users as constant as is possible, but this is of course not actually possible unless IBM sets prices for hardware and software at a constant amount per user. If this were done, it might be an attractive marketing method, provided that the price was low enough. But as things are, there are just too many variables in hardware and software price to keep things constant without literally doing it.

My iDeal iSeries pricing scheme does what the current iSeries line does--charges a little bit more up front for small configurations, and a little bit less per user as hardware and software are added to the machine. For example, a base iDeal iSeries Model 381 with a single 2 GHz PowerPC 970 chip, 1 GB of main memory, and 72 GB of disk could handle about 60 users by my reckoning in a base configuration that costs $27,137 with all users going into the machine using 5250 software. That's about $452 per user for hardware and software. Jumping up to a machine that had four of those 2 GHz processors, 12 GB of main memory, and 540 GB of disk capacity--a machine that I reckon can handle around 712 users after being severely slugged in the manner I described in the first story in this series (I suggested putting in Linux-based print and file servers, adding shadow high availability software, and running the machines with memory bandwidth constrained to keep the price low). In any event, that iDeal iSeries Model 384's hardware would cost $106,420. Adding in OS/400 and DB2/400 with full 5250 access for those 712 users drives the price up to $273,220, or about $384 per user.

I could spend a lot of time building comparison tables between the current iSeries line and the iDeal iSeries, but I don't have that much time this week (I will, however, spend the time this week and get you a fuller comparison to the iSeries line for next week's issue.) One litmus test to how the iDeal iSeries machines compare to the iSeries line is to contrast my pricing with the Green Streak promotion on the Model 270-2342 and the Model 820-2436.

The base Green Streak Model 270-2432 is rated at 1,070 CPW, but has its green screen power slugged down to 50 CPWs using the feature 1519 interactive card. With 1 GB of main memory and four 17.54 GB disk drives, the Model 270 has a list price of $57,624 and a Green Streak price of $28,812. These prices including a 30GB 1/4-inch cartridge tape drive. Using my rule of thumb on CPW ratings and the users it implies is a little weird here. A CPW rating of 1,070 implies a base 54 users and a maximum of about 214 users based on accessing databases other than through the 5250 interface. That maximum user level would obviously require a lot of memory and disks to make it happen. And on the 5250 side of the box, a 50 CPW rating means only between 3 and 10 users can actually work at the machine before it starts running out of gas. Assuming this is correct, that means that the base Model 270 Green Streak machine costs about $9,604 per 5250 user and about $534 per ODBC/JDBC user in a base configuration. Take Green Streak away, and this machine costs twice as much. If you build this machine out to a heftier configuration--say to 3 GB of main memory and ten 36 GB disks--the Model 270 Green Streak would cost $83,692 at list price and $41,846 at Green Streak prices. Assuming that the full number of implied users could get on this machine, based on my CPW rule of thumb, this heavier Model 270 would cost $4,185 for 10 green screen users and about $196 per ODBC/JDBC user if all 214 users that could come in that way did so exclusive of the 5250 protocol.

The closest base configuration in my iDeal iSeries line to the Model 270 Green Streak machine is a Model 381 with a 1.8 GHz PowerPC 970 chip, 1 GB of main memory, and 72 GB of disk costs $25,307, or $469 per user, with 5250 software access for 54 users and $15,857, or $294 per user, with only ODBC/JDBC access for all 54 users. For a heavier configuration of the iDeal iSeries (with 3 GB of memory and 360 GB of disk), I have hardware costs of $41,492. I think that machine can support 216 users, tops, the way I have slugged it. With only ODBC/JDBC users on the machine, the iDeal iSeries 381 as configured above costs $57,692 or $267 per user. If all users come in over 5250 to access the database, then the cost is $95,492, or $442 per user.

IBM's Model 820 Green Streak machine is an 820-2436 configured with 1 GB of main memory, six disk drives with 88 GB of capacity, and a 30GB tape drive. It has a list price of $133,932, and a Green Streak price of $66,966. This machine is rated at 1,100 CPW for server workloads, almost the same as the iSeries Model 270-2432 in the Green Streak deal. The difference is that the Model 820-2436 is rated at 70 CPW on 5250 applications. Again, the closest machine to the Model 820 Green Streak machine in my iDeal iSeries line is the Model 381, the same one that was compared to the iSeries Model 270 Green Streak machine. (However, the Model 831, which I rate at about 1,300 CPWs with a single 1 GHz Power4 processor, could also be compared to it.) The base Model 820 Green Streak machine can support from 4 to 14 green screen users based on my CPW rule of thumb, and from 55 to 220 server-style users. That base machine, with support for four 5250 users, costs $16,742 per green screen user (if all capacity is dedicated to that workload) and $1,218 per ODBC/JBBC (again, assuming all capacity is given to these users).

If you beef up that Model 820 Green Streak machine to support the full complement of green-screen or server workload users--say, give it a total of 4 GB of main memory and 351 GB of disk capacity--it costs $168,308 at list price and $84,154 with the Green Streak discount. That works out to $1,530 per 5250 user (if all users go in through the green screen protocol to the database) and to $383 per ODBC/JDBC user (again, if they all get at data by avoiding the 5250 protocol). That's a bit more expensive than my iDeal iSeries Model 381 for non-5250 workloads and about three times as expensive per user for a machine solely dedicated to green screen users. Double those per user costs after Green Streak goes away on that Model 820.

Suffice it to say, the iDeal iSeries seems to be a better and more fair deal than the Green Streak deal--provided IBM could technically implement such a pricing scheme and had the will to do it. It would take both.

Next week, I'll compare how the iDeal iSeries machines compare generally to the broader iSeries line and also to the Wintel, Lintel, and Unix alternatives outside of the iSeries market. In the final story, I will show what impact a reconfiguration of iSeries technology and marketing that I am talking about might have on the iSeries market.