On Fri, 22 Sep 2006 17:09:50 -0600, Paul Gilmartin <[EMAIL PROTECTED]> wrote:
>Compare: > >Case A: > >o One LPAR with 100 TSO users > >o One LPAR with 50 production jobs > >(Adjust numbers to reality.) versus: > >Case B: > >o Two LPARs, each with 50 TSO users and 25 production jobs. > >OK. You're claiming Case A is the better design. But why? >I had always believed the heterogeneous mix is better because >jobs of each genus can exploit resources little used by the >other(s). I agree with you, Gil. But the best design (assuming that both LPARS run on the same CEC) is to combine them into one LPAR. >I understand, of course, that PR/SM can shift many >resources among partitions according to demand; the partitions >are porous. But not nearly as quickly as WLM can shift resources among address spaces. >What are the exceptions? Main memory? WLM >profiles? Other (specify)? Of course TSO jobs should have >dispatching profiles different from batch -- shorter CPU time >slices for TSO, etc. But couldn't that be done by assigning >different profiles to the TSO jobs, even within the same >LPAR. Yes. But in WLM, we don't specify the technical details of how to divide resources. Rather, we specify goals. For example, we might specify 90th percentile period 1 TSO complete within 0.5 second. Or that 50% of production batch jobs complete within 30 minutes. All jobs in a given service class are treated the same, so that in the production batch example, WLM gives the same dispatching priority, etc. to all production batch jobs. The goal helps WLM to prioritize the work and to decide what initiators to start. >Perhaps it's the design of WLM that doesn't account >for profoundly diverse behaviors within a single LPAR. Ahh, but it does. Tom Marchant ---------------------------------------------------------------------- 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
