Обсуждение: Indexes
Does anyone think it would be a good idea, or is it even practical, to have a 'indx' subdirectory along side of the 'base' directory? I was thinking that, if it were an easy modification, that it could be an easy way to separate data and indexes to different hard disks.
>>>mlw said:> Does anyone think it would be a good idea, or is it even practical, to > have a 'indx' subdirectory along sideof the 'base' directory?> > I was thinking that, if it were an easy modification, that it could be > an easy way to separatedata and indexes to different hard disks. My guess is that having indexes at separate location on a per-databases basis would give greater benefit. Perhaps this couldbe extended to all databases as well, but per-database location should be better? Or do you think of something like data/base/xxxxx/indx where xxxxx is the database oid. Daniel
> mlw said: >>> Does anyone think it would be a good idea, or is it even practical, to >>> have a 'indx' subdirectory along side of the 'base' directory? >>> >>> I was thinking that, if it were an easy modification, that it could be >>> an easy way to separate data and indexes to different hard disks. This and other quick hacks have been discussed before. I think the consensus has been to do nothing until someone gets around to writing a general-purpose tablespace implementation. FWIW, I don't think it would be an easy modification. The low-level file access code doesn't even know whether a given relation is an index or not (... and I don't think it should know). By the time you get done implementing something reasonable at that level, you've got the infrastructure you need for tablespaces. regards, tom lane
On Thu, 16 Jan 2003, Tom Lane wrote: > > mlw said: > >>> Does anyone think it would be a good idea, or is it even practical, to > >>> have a 'indx' subdirectory along side of the 'base' directory? > >>> > >>> I was thinking that, if it were an easy modification, that it could be > >>> an easy way to separate data and indexes to different hard disks. > > This and other quick hacks have been discussed before. I think the > consensus has been to do nothing until someone gets around to writing > a general-purpose tablespace implementation. Wasn't someone just about done with a tablespace implementation? I certainly remember some discussion on this subject a few months ago. -- Nigel J. Andrews