IDS 7.31 PSORT_DBTEMP

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi All, I am a little confused about setting the PSORT_DBTEMP env. variable... Most of the docs state that setting this variable is solely for parallel sorting.. Now if this variable is not set, will it default to DBSPACETEMP ? Or will there be no benefit for parallel sorting at all ? Plus can one define dbspaces rather than files for PSORT_DBTEMP ? thanks, Peter -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.6 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFC/IMZpTld8yX8i1kRAmTMAKDMKpyFeznoz+TegHka4vqBbMBl+QCgrxty zAsrAJRCMAr9FNxj+xBOXsc= =gzNk -----END PGP SIGNATURE-----

----- Original Message ----- From: Peter Van E.... <Peter.VAN.ECK@dhl.com> At: 8/12 9:37 > Hi All, Hello. > I am a little confused about setting the PSORT_DBTEMP env. variable... > Most of the docs state that setting this variable is solely for > parallel sorting.. Correct. Though that means it is also used for parallel sorts during index builds (online not with oncheck) and update statistics processing. > Now if this variable is not set, will it default to DBSPACETEMP ? Or > will there be no benefit for parallel sorting at all ? Yes, as long as PSORT_NPROCS is >1 and PDQPRIORITY is >=2 (PDQPRIORITY=1 only enables parallel scans). If parallel sorts are enabled using PSORT_NPROCS and PDQPRIORITY and PSORT_DBTEMP is not set then the temp dbspaces listed in DBSPACETEMP (but not 'normal' dbspaces listed there) will be used if a sort needs to write sort-work files to disk (MGM settings determine how much data is sorted in memory before going to disk - see the Performance Guide). > Plus can one define dbspaces rather than files for PSORT_DBTEMP ? > thanks, No, and you do not want to. Dbspaces are normally RAW devices and are written using O_SNC if they are COOKED devices or filesystem files bypassing the system cache. One advantage of using PSORT_DBTEMP is that the filesystems listed ARE using the system cache. Since sort-work files are short lived in general, they will tend to remain completely in cache and may never be written to disk before IDS deletes them so they are much lower overhead and resource users than using dbspaces for sort-work files. If you list from hree to six separate filesystems in PSORT_DBTEMP you will optimize sort times according to all of my testing. You need at least three because of the way the files are created and reread to avoid any contention if the IO does go physically to disk. Art S. Kagel > Peter