To: Mary Cluney who wrote (138744 ) 7/6/2001 9:06:47 PM From: pgerassi Read Replies (1) | Respond to of 186894 Dear Mary: Unlike you, I have written software for Medical billing and other such tasks. The amounts of CPU power required are quite small compared to workstations that simulate processes and devices. What is required is large databases (read storage). One such device is the tape library. It holds thousands of tapes yet the robotic systems are controlled by a simple PC system (with specialized software and interfaces of course). Most storage server types do not need the high end PC CPUs (the reign of P3 Xeons of 700 MHz and below bear this out). What they need is IO power (disk drives, a tape drive (for backup), and HS LAN ports). Stuff like 12 SCSI drives (4 on each of three channels) costing upwards of $12K, $2K of controllers, $2K for memory, $2K for the motherboard and the chassis (hey this is industrial 24-7 hardware we are talking about here). The CPUs are a pair of P3 600MHz ones. For I/O bound tasks, they are more than adaquate. So $200 to $300 for CPUs and $20K for the rest. Add $2K for the OS and $8K for the RDBMS and we are back to the 1% range. For typical server apps where they serve out storage, RDBMS power, etc., are all I/O bound tasks. In these areas, CPUs are a small percentage of the cost. For compute bound tasks, then you get the larger CPU percentages. These are typical scientific and engineering tasks. Do not make the mistake of confusing the two. In the $60B server market, the former areas make the bulk of the revenues say, 80 to 90%. The reason you see the later a lot is that is what is "sexy" about the upper range like the Supercomputers and MPPs. This has always been the way of it. Those large businesses have the "sea of disk" which was a football field sized room filled with washing machine disk drives and tape machines each costing $100K (like 2,400 of them) attached to mainframes costing $10-20M. Even when number crunching was king like oil exploration, the CPUs were like 2% to 3% of the overall cost of the number crunching behemoths. Those machines had like 480GB of attached disk. A simple PC tower can have more than that now and still compute faster than that oil exploration business running system of the past for 0.01% of the cost. What has changed in the paradigm is that the software costs far exceed the hardware costs. In those old days it was exactly opposite. When you buy a "turn key" server from a VAR, you are paying mostly for software and services. The hardware is a small percentage of those selling prices. A typical rule of thumb is 10% of the contract price is hardware related. If that is the basis of the $60B in server revenue, your estimates are way too high for the CPU component portion. Mine were for all hardware that goes into and attaches to the server (like what is in the cabinets containing the server (this is typical of what upper management views as the "server")). My only question was if the RDBMS software would be included like the OS is included in that figure you specified. Your theory assumes that the server is what would be called the bare bones kit in the PC world (the case, MB, CPUs, memory, and HSFs). I think that it must be one of the first two. The middle one gives you the benefit of the doubt, but the former is more likely the case. Just take the way in which network attached anything is called nowadays. Print servers (NAS printers), scan servers, terminal servers, modem servers (used to be called modem banks), storage servers, tape servers, CD-ROM servers, backup servers, boot servers, and so forth. I'm sure you can think of more. Pete