Re: Perfomance Tuning - Mailing list pgsql-performance

On Tue, 2003-08-12 at 13:39, Bruce Momjian wrote:
> OK, I got some hard evidence.  Here is a discussion on the Linux kernel
> mailing list with postings from Allen Cox (ac Linux kernels) and Stephen
> Tweedie (ext3 author).
>
>     http://www.tux.org/hypermail/linux-kernel/1999week14/subject.html#start
>
> Search for "softupdates and ext2".
>
> Here is the original email in the thread:
>
>     http://www.tux.org/hypermail/linux-kernel/1999week14/0498.html
>
> Summary is at:
>
>     http://www.tux.org/hypermail/linux-kernel/1999week14/0571.html
>
> and conclusion in:
>
>     http://www.tux.org/hypermail/linux-kernel/1999week14/0504.html
>
> I now remember the issue --- ext2 makes all disk changes asynchonously
> (unless you mount it via sync, which is slow).  This means that the file
> system isn't always consistent on disk.
>
> UFS has always sync metadata (file/directory creation) to the disk so
> the disk was always consistent, but doesn't sync the data to the disk,
> for performance reasons.  With soft updates, the metadata writes are
> delayed, and written to disk in an order that keeps the file system
> consistent.
>
> Is this enough evidence, or should I keep researching?

This is all 4 years old, though.  Isn't that why the ext3 "layer" was
created, and filesystems like reiserFS, XFS and (kinda) JFS were added
to Linux?

> ---------------------------------------------------------------------------
>
> Neil Conway wrote:
> > On Tue, Aug 12, 2003 at 12:52:46AM -0400, Bruce Momjian wrote:
> >  I don't use Linux and was just repeating what I had heard from others,
> > > and read in postings.  I don't have any first-hand experience with ext2
> > > (except for a laptop I borrowed that wouldn't boot after being shut
> > > off), but others on this mailing list have said the same thing.
> >
> > Right, and I understand the need to answer users asking about
> > which filesystem to use, but I'd be cautious of bad-mouthing
> > another OSS project without any hard evidence to back up our
> > claim (of course if we have such evidence, then fine -- I
> > just haven't seen it). It would be like $SOME_LARGE_OSS
> > project saying "Don't use our project with PostgreSQL, as
> > foo@bar.org had data corruption with PostgreSQL 6.3 on
> > UnixWare" -- kind of annoying, right?
> >
> > > > (a) ext3 does metadata-only journalling by default
> > >
> > > If that is true, why was I told people have to mount their ext3 file
> > > systems with metadata-only.  Again, I have no experience myself, but why
> > > are people telling me this?
> >
> > Perhaps they were suggesting that people mount ext2 using
> > data=writeback, rather than the default of data=ordered.
> >
> > BTW, I've heard from a couple different people that using
> > ext3 with data=journalled (i.e. enabling journalling of both
> > data and metadata) actually makes PostgreSQL faster, as
> > it means that ext3 can skip PostgreSQL's fsync request
> > since ext3's log is flushed to disk already. I haven't
> > tested this myself, however.
> >
> > -Neil

--
+---------------------------------------------------------------+
| Ron Johnson, Jr.        Home: ron.l.johnson@cox.net           |
| Jefferson, LA  USA                                            |
|                                                               |
| "Man, I'm pretty.  Hoo Hah!"                                  |
|    Johnny Bravo                                               |
+---------------------------------------------------------------+