Обсуждение: FW: XFS filessystem for Datawarehousing -2
Sorry, forgot to ask: What is the recommended/best PG block size for DWH database? 16k, 32k, 64k ? What hsould be the relation between XFS/RAID stripe size and PG block size ? Best Regards. Milen Kulev -----Original Message----- From: pgsql-performance-owner@postgresql.org [mailto:pgsql-performance-owner@postgresql.org] On Behalf Of Milen Kulev Sent: Tuesday, August 01, 2006 11:50 PM To: pgsql-performance@postgresql.org Subject: [PERFORM] XFS filessystem for Datawarehousing I intend to test Postgres/Bizgres for DWH use. I want to use XFS filesystem to get the best possible performance at FS level(correct me if I am wrong !). Is anyone using XFS for storing/retrieving relatively large amount of data (~ 200GB)? If yes, what about the performance and stability of XFS. I am especially interested in recommendations about XFS mount options and mkfs.xfs options. My setup will be roughly this: 1) 4 SCSI HDD , 128GB each, 2) RAID 0 on the four SCSI HDD disks using LVM (software RAID) There are two other SATA HDD in the server. Server has 2 physical CPUs (XEON at 3 GHz), 4 Logical CPUs, 8 GB RAM, OS = SLES9 SP3 My questions: 1) Should I place external XFS journal on separate device ? 2) What should be the journal buffer size (logbsize) ? 3) How many journal buffers (logbufs) should I configure ? 4) How many allocations groups (for mkfs.xfs) should I configure 5) Is it wortj settion noatime ? 6) What I/O scheduler(elevators) should I use (massive sequencial reads) 7) What is the ideal stripe unit and width (for a RAID device) ? I will appreciate any options, suggestions, pointers. Best Regards. Milen Kulev ---------------------------(end of broadcast)--------------------------- TIP 4: Have you searched our list archives? http://archives.postgresql.org
Milen, On 8/1/06 3:19 PM, "Milen Kulev" <makulev@gmx.net> wrote: > Sorry, forgot to ask: > What is the recommended/best PG block size for DWH database? 16k, 32k, 64k > ? > What hsould be the relation between XFS/RAID stripe size and PG block size ? We have found that the page size in PG starts to matter only at very high disk performance levels around 1000MB/s. Other posters have talked about maintenance tasks improving in performance, but I haven't seen it. - Luke
I was kinda thinking that making the Block Size configurable at InitDB time would be a nice & simple enhancement for PG 8.3. My own personal rule of thumb for sizing is 8k for OLTP, 16k for mixed use, & 32k for DWH.
I have no personal experience with XFS, but, I've seen numerous internal edb-postgres test results that show that of all file systems... OCFS 2.0 seems to be quite good for PG update intensive apps (especially on 64 bit machines).
On 8/1/06, Luke Lonergan <llonergan@greenplum.com> wrote:
Milen,
On 8/1/06 3:19 PM, "Milen Kulev" <makulev@gmx.net> wrote:
> Sorry, forgot to ask:
> What is the recommended/best PG block size for DWH database? 16k, 32k, 64k
> ?
> What hsould be the relation between XFS/RAID stripe size and PG block size ?
We have found that the page size in PG starts to matter only at very high
disk performance levels around 1000MB/s. Other posters have talked about
maintenance tasks improving in performance, but I haven't seen it.
- Luke
---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?
http://archives.postgresql.org
Hi Dennis,
I am just cusrios to try PG with different block sizes ;) I don't know how much performance the bigger block size will bring (I mean 32k or 64k , for example, for DWH applikations).
I am surprised to hear that OCFS2.0 (or any her FS usind direct I/O) performs well with PG. A month ago I have performed a simple test with Veritas FS, with and than without cache (e.g. direct I/O). I have started 1 , then 2, , then 3, then 4 parallel INSERT processes.
Veritas FS WITH FS cache outperformed the direct I/O version by factor 2-2.5 !
I haven't tested woth OCFS2.0 though. I am not sure that OCFS2.0 is the good choice for PG data and index filesystems.
For WAL -> perhaps.
Best Regards. Milen
-----Original Message-----
From: pgsql-performance-owner@postgresql.org [mailto:pgsql-performance-owner@postgresql.org] On Behalf Of Denis Lussier
Sent: Thursday, August 03, 2006 7:36 AM
To: Luke Lonergan
Cc: Milen Kulev; pgsql-performance@postgresql.org
Subject: Re: [PERFORM] XFS filessystem for Datawarehousing -2
I was kinda thinking that making the Block Size configurable at InitDB time would be a nice & simple enhancement for PG 8.3. My own personal rule of thumb for sizing is 8k for OLTP, 16k for mixed use, & 32k for DWH.
I have no personal experience with XFS, but, I've seen numerous internal edb-postgres test results that show that of all file systems... OCFS 2.0 seems to be quite good for PG update intensive apps (especially on 64 bit machines).On 8/1/06, Luke Lonergan <llonergan@greenplum.com> wrote:Milen,
On 8/1/06 3:19 PM, "Milen Kulev" <makulev@gmx.net> wrote:
> Sorry, forgot to ask:
> What is the recommended/best PG block size for DWH database? 16k, 32k, 64k
> ?
> What hsould be the relation between XFS/RAID stripe size and PG block size ?
We have found that the page size in PG starts to matter only at very high
disk performance levels around 1000MB/s. Other posters have talked about
maintenance tasks improving in performance, but I haven't seen it.
- Luke
---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?
http://archives.postgresql.org
denisl@enterprisedb.com ("Denis Lussier") writes: > I have no personal experience with XFS, but, I've seen numerous > internal edb-postgres test results that show that of all file > systems... OCFS 2.0 seems to be quite good for PG update intensive > apps (especially on 64 bit machines). I have been curious about OCFS for some time; it sounded like a case where there could possibly be some useful semantic changes to filesystem functionality, notably that: - atime is pretty irrelevant; - it might try to work with pretty fixed block sizes (8K, perhaps?) rather than try to be efficient at handling tiny files It sounds like it ought to be able to be a good fit. Of course, with a big warning sticker of "what is required for Oracle to work properly is implemented, anything more is not a guarantee" on it, who's going to trust it? -- select 'cbbrowne' || '@' || 'cbbrowne.com'; http://www.ntlug.org/~cbbrowne/oses.html "There isn't any reason why Linux can't be implemented as an enterprise computing solution. Find out what you've been missing while you've been rebooting Windows NT." - Infoworld
I agree that OCFS 2.0 is NOT a general purpose PG (or any other) solution. My recollection is that OCFS gave about 15% performance improvements (same as setting some aggressive switches on ext3). I assume OCFS has excellent crash safety with its default settings but we did not test this as of yet. OCFS now ships as one of the optional FS's that ship with Suse. That takes care of some of the FUD created by Oracle's disclaimer below.
OCFS 2 is much more POSIX compliant than OCFS 1. The BenchmarkSQL, DBT2, & Regression tests we ran on OCFS 2 all worked well. The lack of full Posix compliance did cause some problems for configuring PITR.
--Denis http://www.enterprisedb.com
On 8/3/06, Chris Browne <cbbrowne@acm.org > wrote:
Of course, with a big warning sticker of "what is required for Oracle
to work properly is implemented, anything more is not a guarantee" on
it, who's going to trust it?
--