On 29 Sep 2002 at 0:43, Justin Clift wrote: > Shridhar Daithankar wrote: > The reason that I was thinking of having a different path per index > would be for high volume situations like this: > > /dev/dsk1 : /pgdata <- data here > /dev/dsk2 : /pgindexes1 <- some indexes here > /dev/dsk3 : /pgindexes2 <- some ultra-high volume activity here
I would say this would look better.. /pgdata -indexes --index1 ---indexfiles --index2 ---indexfiles Where index1 and index2 are two different indexes. Just like each table gets it's own directory, each index gets it's own directory as well. So the admin would/can tune on per object basis rather than worrying about creating right group of objects and then tuning about that group. If required throwing per database transaction log there as well might prove a good idea. It would insulate one db from load of other, as far as I/O is concerned.. This possiblity is not lost with this scheme but it just gets something simpler IMO.. Just illustration of my another post on hackers on this topic.. Bye Shridhar -- You're too beautiful to ignore. Too much woman. -- Kirk to Yeoman Rand, "The Enemy Within", stardate unknown ---------------------------(end of broadcast)--------------------------- TIP 4: Don't 'kill -9' the postmaster
