Greetings Mark Thomen and all, I'm in the process of tuning our catalogs. In looking at the results of a F CATALOG,REPORT,PERFORMANCE command I see a huge difference in the amount of time to process a BCS ENQ between systems in a GRS-plex and those that are not in a GRS-plex.
With the two systems I have in the GRS-plex I see ENQ and DEQ overhead in the 6-8 msec range such as these values on one of those systems: -----CATALOG EVENT---- --COUNT-- ---AVERAGE--- Entries to Catalog 2,347K 36.096 MSEC BCS ENQ Shr Sys 4,024K 6.941 MSEC BCS ENQ Excl Sys 76,131 6.868 MSEC BCS DEQ 6,080K 6.638 MSEC However on another production system that is not part of a GRS-plex, and also on a sandbox system that is not part of any GRS-plex, I see ENQ and DEQ overheads significantly smaller: -----CATALOG EVENT---- --COUNT-- ---AVERAGE--- Entries to Catalog 1,540K 7.247 MSEC BCS ENQ Shr Sys 2,698K 0.107 MSEC BCS ENQ Excl Sys 119,786 0.115 MSEC BCS DEQ 4,256K 0.038 MSEC In an attempt to reduce that overhead I made a change to the Shareoptions of the user catalogs. ALL of our catalogs were originally defined with SHR (3,4). I changed those that were not shared to SHR(3,3) on the development system that is part of the GRS-plex in the hopes that this would reduce the overhead. We only have 1 user catalog that is shared between the systems and that one is still defined as SHR(3,4). I and IPL'ed after the change, yet I'm not seeing any appreciable difference in the ENQ and DEQ overheads. I did notice that there were now extra reported lines to make me believe the change has had some affect: -----CATALOG EVENT---- --COUNT-- ---AVERAGE--- Entries to Catalog 170,669 39.737 MSEC BCS ENQ Shr 255,768 7.626 MSEC BCS ENQ Shr Sys 25,688 7.707 MSEC BCS ENQ Excl 5,537 7.701 MSEC BCS ENQ Excl Sys 11 10.456 MSEC BCS DEQ 435,086 7.706 MSEC I also notice the VVDS I/O has reduced significantly. However, I was hoping to have seen a tremendous drop in the ENQ/DEQ overhead. Am I missing something? I've checked the performance recommendations in Item II10752. *) We are a two system GRS RING with RESMIL(1) specified. *) I increased STRNO from 3 to 7 for all catalogs. *) I have RNLDEF RNL(CON) TYPE(GENERIC) QNAME(SYSIGGV2) *) I have RNLDEF RNL(EXCL) TYPE(GENERIC) QNAME(SYSZVVDS) *) I'm Z/OS 1.4 and don't have a coupling facility. The only other thing I'm considering at the moment is that all of the DASD is defined as SHARED even though only the volume with the shared catalog is online to both systems. All other DASD is offline to the other system in the GRS-plex. I would have thought that by changing to shareoptions(3,3) that I shouldn't need to change the DASD to unshared. Currently all the usercatalogs are using ISC cache. I haven't tried VLF yet. But I can't see that should make a difference to the ENQ/DEQ costs. Does anyone have any experience with going to/from a GRS-plex or have systems that are plexed and those that are not? Do you see the same differences in overhead? Should I have expected to see a reduction in overhead changing shareoptions to (3,3)? Thanks, Tom Rusnak ---------------------------------------------------------------------- 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
