Steve Thompson writes: >Price of CPU is determined by cost of processor (the whole box), all the >software that runs on it, all the power to run the system (CEC, RAID, >ATL, etc.), and then costs of personnel to run the system.
With you so far. That's called total price, assuming you've got all the prices. Add things like depreciation, interest rates, and various other prices over time and you've got "costs." >This is then divided by the number of hours per day the system is in use. ...And you lost me there! :-) What you're describing is the *average* (or mean) price per CPU-hour. (Divide by seconds to get CPU-seconds.) Most importantly, if you set your chargebacks based on this number you WILL have VERY serious user behavioral problems. Actually, costs are strongly curved (and not a line), and the curve intercepts the Y axis FAR from zero. There are fixed costs to any IT infrastructure. The incremental costs -- the *marginal* costs -- are very, very different from the average costs at any point on the curve. It's the marginal costs that are the appropriate ones for chargeback. If you do use chargebacks, and you don't understand the difference between average and marginal costs, you're in BIG trouble. >And then there are a few other considerations for normalization (how >many years is the system expected to be kept, are their shift >differentials (example: for work run from Midnight to 7AM) and so >forth). There are some other big ones. Here's a partial list: * What WLM service class does the work enjoy? If it's completely discretionary workload below a cap, it's (basically) "free." Does it run at peak or off-peak? * Is the workload zAAP eligible? zIIP eligible? zAAPs and zIIPs are different from CPs. * In which LPAR is the workload running? Different LPARs have different costs. They often vary in the composition of sub-capacity eligible software, for example. * Is the workload using Parallel Sysplex services or not? What sort of disaster recovery protection does the workload enjoy? * How much memory is the workload using? How much storage (space and IOPS)? How much crypto? How much network I/O? * What software product(s) does the workload require? Is the workload the sole consumer of a particular pricey tool, for example? * Any printing? * Is it workload running on CBU (Capacity Backup)? CBU is for your declared disasters, or for disaster rehearsals, and has a very different price. Same idea with workload running on MSU-day hardware (On-Off Capacity On Demand). I've also seen cases where IT accountants lump 100% of the data center costs (floor space, power, cooling) into the mainframe budget even though the mainframe is the smallest, coolest, and lowest electric consumer in the data center. And there are many, many cases where chargebacks are levied via the mainframe simply because the mainframe has SMF and the other systems don't. And how do you measure the price of IFLs? If it's an IFL-only system it's a little easier, but what if it isn't? Eric Bielefeld writes: >Well, actually you can purchase a fraction of a CPU. The z9 BC, z890, >and z800 all have many settings for their CPU speed. But, on the larger >processors you can't do that. You can't add one CPU at half the speed of >the others - they all have to be the same speed. The System z9 EC does have sub-capacity configurations below 8 full speed processors. The 4xx, 5xx, and 6xx models in the 2094 series are all sub-capacity (fractional CP) models, ranging from the 401 to the 608. But Eric raises a good point, to restate it a bit differently. The price curve actually has a little bit of a "sawtooth" pattern to it. In the past -- or if you're still running in the past or pretending it's the past, e.g. OS/390 -- the "teeth" were big, sharp, and jagged. Now the curve is much smoother with both sub-capacity hardware and sub-capacity software being so popular and prevalent. You can buy as little as one MSU-day of IBM OTC software, one MSU-month of IBM MLC software, and tiny MSU-day increments of hardware (as Eric points out). You can also buy IT staffing by the hour much more easily than in the past. In fact, one of the "lumpiest" IT costs nowadays is when data center managers have to bust down a wall and/or install new power/cooling systems because they filled their data centers with distributed servers. That means that next Dell server costs $10M. :-) Obviously it's not easy to take all these factors (and others) into account, then do the same for other IT infrastructure. It's part science and part art. But if your organization isn't doing at least a "reasonable" job understanding and allocating costs -- if there's a lack of understanding between average and marginal, for example -- there are going to be serious challenges making correct decisions. And the ultimate "penalty" is that the business managers throw up their arms and outsource everything. Much of the value outsourcers bring is that they understand costs better than others more often than not. Of course the outsourcers charge a profit for their services, and they perform to written agreements. If it's not written down, the outsourcer doesn't have to do it. - - - - - Timothy Sipples IBM Consulting Enterprise Software Architect Specializing in Software Architectures Related to System z Based in Tokyo, Serving IBM Japan and IBM Asia-Pacific E-Mail: [EMAIL PROTECTED] ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
