Re: [PERFORM] Filesystems WAS: Perfomance Tuning
On Tue, Aug 12, 2003 at 04:31:19PM -0400, Richard Welty wrote: > impractical. has any consideration been given to providing a way to manage > such a deployment? Plenty. No-one's completed an implementation yet. > or is it the judgement of the hackers community that a monsterous raid-10 > array offers comparable performance? It's tough to say, but I _can_ tell you that, so far in my tests, I've never been able to prove an advantage in separating even the WAL on a Sun A5200. That's not a result yet, of course, just a bit of gossip. A -- Andrew Sullivan 204-4141 Yonge Street Liberty RMS Toronto, Ontario Canada <[EMAIL PROTECTED]> M2P 2A8 +1 416 646 3304 x110 ---(end of broadcast)--- TIP 5: Have you checked our extensive FAQ? http://www.postgresql.org/docs/faqs/FAQ.html
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
On Tue, 12 Aug 2003, Bruce Momjian wrote: > > I think Gavin Sherry is working on this. I am CC'ing him. > > --- Yes I am working on this. I am about 50% of the way through the patch but have been held up with other work. For those who are interested, it basically allow: 1) creation of different 'storage' locations. Tables and indexes can be created in different storage locations. Storage locations can also be assigned to schemas and databases. Tables and indexes will default to the schema storage location if STORAGE 'store name' is not provided to CREATE This will cascade to the default database storage location if the schema was not created with STORAGE 'store name'. 2) the patch will allow different storage locations to have different rand_cost parameters passed to the planner. 3) the patch *will not* address issues concerning quotas, resource management, WAL/clog, temp or sort spaces. Will keep everyone posted if/when I finish. Thanks, Gavin ---(end of broadcast)--- TIP 2: you can get off all lists at once with the unregister command (send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
I think Gavin Sherry is working on this. I am CC'ing him. --- Christopher Kings-Lynne wrote: > > there are advantages to being able to split the database across a slew of > > disk drives. if we accept the notion of using the native OS filesystem on > > each, it would seem that being able to direct various tables and indices > to > > specific drives might be a valuble capability. i know that i could go into > > /var/lib/pgsql/data/base and fan the contents out, but this is unweildy > and > > impractical. has any consideration been given to providing a way to manage > > such a deployment? > > We've got a little bunch of us tinkering with a tablespace implementation. > However, it's been staller for a while now. > > Chris > > > ---(end of broadcast)--- > TIP 7: don't forget to increase your free space map settings > -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 359-1001 + If your life is a hard drive, | 13 Roberts Road + Christ can be your backup.| Newtown Square, Pennsylvania 19073 ---(end of broadcast)--- TIP 6: Have you searched our list archives? http://archives.postgresql.org
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
On Wed, 13 Aug 2003 09:48:18 +0800 Christopher Kings-Lynne <[EMAIL PROTECTED]> wrote: > We've got a little bunch of us tinkering with a tablespace > implementation. > However, it's been staller for a while now. interesting. i'm involved in the very early stages of a startup that is likely to do a prototype using Java and PostgreSQL. tablespace and replication are issues that would weigh heavily in a decision to stick with PostgreSQL after the prototype. richard -- Richard Welty [EMAIL PROTECTED] Averill Park Networking 518-573-7592 Java, PHP, PostgreSQL, Unix, Linux, IP Network Engineering, Security ---(end of broadcast)--- TIP 9: the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
Greg, > FWIW, Informix can be run using a "cooked" (Unix) file for storing data or > it uses "raw" disk space and bypasses the ordinary (high level) UNIX > controllers and does its own reads/writes. About 10 times faster and safer. > Of course, itmay have taken a lot of programmer time to make that solid. > But the performance gains are significant. Yes, but it's still slower than PostgreSQL on medium-end hardware. ;-) This idea has been discussed numerous times on the HACKERS list, and is a (pretty much) closed issue. While Oracle and SQL Server use their own filesystems, PostgreSQL will not because: 1) It would interfere with our cross-platform compatibility. PostgreSQL runs on something like 20 OSes. 2) The filesystem projects out there are (mostly) well-staffed and are constantly advancing using specialized technology and theory. There's no way that the PostgreSQL team can do a better job in our "spare time". 3) Development of our "own" filesystem would then require PostgreSQL to create and maintain a whole hardware compatibility library, and troubleshoot problems on exotic hardware and wierd RAID configurations. 4) A database FS also often causes side-effect problems; for example, one cannot move or copy a SQL Server partition without destroying it. Of course, that could all change if some corp with deep pockets steps in an decides to create a "postgresFS" and funds and staffs the effort 100%. But it's unlikely to be a priority for the existing development team any time in the forseeable future. -- Josh Berkus Aglio Database Solutions San Francisco ---(end of broadcast)--- TIP 3: if posting/reading through Usenet, please send an appropriate subscribe-nomail command to [EMAIL PROTECTED] so that your message can get through to the mailing list cleanly
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
On Tue, 12 Aug 2003 13:09:42 -0700 Josh Berkus <[EMAIL PROTECTED]> wrote: > This idea has been discussed numerous times on the HACKERS list, and is > a > (pretty much) closed issue. While Oracle and SQL Server use their own > filesystems, PostgreSQL will not because: ... > 2) The filesystem projects out there are (mostly) well-staffed and are > constantly advancing using specialized technology and theory. There's > no way > that the PostgreSQL team can do a better job in our "spare time". i consider this a fair answer, but i have a slightly different question to ask, inspired by my discussions with a good friend who is a top notch Informix DBA. there are advantages to being able to split the database across a slew of disk drives. if we accept the notion of using the native OS filesystem on each, it would seem that being able to direct various tables and indices to specific drives might be a valuble capability. i know that i could go into /var/lib/pgsql/data/base and fan the contents out, but this is unweildy and impractical. has any consideration been given to providing a way to manage such a deployment? or is it the judgement of the hackers community that a monsterous raid-10 array offers comparable performance? i forget how large the data store on my friend's current project is, but i'll check. knowing the size and transaction rate he's dealing with might put a finer point on this discussion. richard -- Richard Welty [EMAIL PROTECTED] Averill Park Networking 518-573-7592 Java, PHP, PostgreSQL, Unix, Linux, IP Network Engineering, Security ---(end of broadcast)--- TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED]
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
> there are advantages to being able to split the database across a slew of > disk drives. if we accept the notion of using the native OS filesystem on > each, it would seem that being able to direct various tables and indices to > specific drives might be a valuble capability. i know that i could go into > /var/lib/pgsql/data/base and fan the contents out, but this is unweildy and > impractical. has any consideration been given to providing a way to manage > such a deployment? We've got a little bunch of us tinkering with a tablespace implementation. However, it's been staller for a while now. Chris ---(end of broadcast)--- TIP 7: don't forget to increase your free space map settings
Re: [PERFORM] Filesystems WAS: Perfomance Tuning
> specific drives might be a valuble capability. i know that i could go into > /var/lib/pgsql/data/base and fan the contents out, but this is unweildy and > impractical. has any consideration been given to providing a way to manage > such a deployment? The ability to take various database objects and store them in different locations, sometimes referred to as table spaces, will probably be done in the future. There was a substantial group not all that long ago that was organizing to complete the implementation. > or is it the judgement of the hackers community that a monsterous raid-10 > array offers comparable performance? Often performs well enough... But a raid-10 for data, a stripe for indexes, and a mirror for WAL will offer better performance :) signature.asc Description: This is a digitally signed message part