Not running IRLM at a higher dispatching priority runs counter to IBM's advice. Before goal mode, DB2 would write a warning message when IRLM was not at a higher DP than the other DB2 address spaces.
The reason for this is that IRLM runs in SRB mode (asynchronously) as opposed to cross-memory mode (synchonously). Just take a look at any IRLM under RMF and you will see the SRB time. If it were XMEM time, it would accrue to the requester's address space, not IRLM's. The reason it makes sense is that by virtue of z/systems architecture SRBs ALWAYS run at a higher priority than TCBs. IRLM must service ALL lock requests as a common service at the highest priority. Lock management cannot tolerate running synchronously at the low priority of some Johnny-come-lately unit of work. In fact, z/OS will raise the DP of lock/latch holders to avoid contending with lock/latch waiters, but the actual locking operation runs at the priority of IRLM. The reason you get away with setting it at the same priority as the other DB2 address spaces is that the operations are themselves miniscule compared to the CPU consumption of the other address spaces. Furthermore, while a task is using DBM1 services in XMEM mode, the housekeeping overhead and I/O management is largely performed under SRB mode in DBM1 at the priority of DBM1. As a point of reference, see Address space communication at http://publib.boulder.ibm.com/infocenter/zoslnctr/v1r7/index.jsp?topic=/ com.ibm.zconcepts.doc/zconcepts_84.html. It is true that dispatching priorities cannot be controlled directly in goal mode, however SYSTEM and SYSSTC are by design the highest fixed dispatching priorities in the system. When the requester and provider of service are separate entities, as they are in this case, common sense dictates the provider run at a higher priority than the requester. db -----Original Message----- From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On Behalf Of Ted MacNEIL Sent: Tuesday, July 03, 2007 8:43 AM To: IBM-MAIN@BAMA.UA.EDU Subject: Re: WLM Question >Actually I feel your IRLM should run in SYSSTC. Every request is going >to go through that. You don't want to take a chance on it waiting for >some TSO user for CPU. Actually, that's what I said in my post (... all three address spaces in SYSSTC ...). Also, what I said was that it all runs through cross-memory services, so it will run at the priority of the requesting task -- NOT at the IRLM priority. Take a look at any IRLM under RMF (or your monitor of choice). Except for the interval it actually starts, you will see no TCB time consumed. Just SRB. Why? Cross-Memory Services (the [very] little TCB time was consumed starting the IRLM ASID). - Too busy driving to stop for gas! ---------------------------------------------------------------------- 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 ---------------------------------------------------------------------- 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
