Thread: Re: Which hardware/filesystem for postgresql?
In the last exciting episode, Cristian Veronesi <c.veronesi@crpa.it> wrote:
> Hello, my company is starting to propose postgresql-based solutions
> to our clients. Our recommended operating system is SuSE
> Linux. Which disk architecture should we recommend for postgresql
> servers? I was thinking about RAID10.
That all depends on what disk hardware there is, and what you're
doing. The reason why they have all of the _various_ RAID levels,
instead of just 1, is that flexibility is often needed when hardware
varies and when the meaning of "best" varies.
> Also, what Linux filesystem should we use? I was thinking about XFS.
> Which filesystems are you using?
The best performance results I have seen on Linux systems have
involved the use of JFS. I found XFS to be a little slower, and it
has the distinct demerit that it is not in the 'official' kernel tree
yet, thereby meaning that you have to get into the pain of managing
heavily-patched kernels. The "kernel management" issue strikes me as
being a much bigger deal than the relatively minor performance
difference.
--
output = reverse("ac.notelrac.teneerf" "@" "454aa")
http://cbbrowne.com/info/sgml.html
"Wintel: A Wasteland of Useless Software - If the bazillions of
programs out there actually amount to something, why is everyone using
MICROS~1 Office, Word, Excel, PowerPoint, Access, ..."
-- cbbrowne@hex.net
Cris Carampa <cris119@operamail.com> writes:
> Christopher Browne wrote:
>
>> The best performance results I have seen on Linux systems have
>> involved the use of JFS. I found XFS to be a little slower, and it
>> has the distinct demerit that it is not in the 'official' kernel tree
>> yet, thereby meaning that you have to get into the pain of managing
>> heavily-patched kernels. The "kernel management" issue strikes me as
>> being a much bigger deal than the relatively minor performance
>> difference.
>
> Thank you for your answer (it's still me, now I'm using my "official"
> usenet account :))
You may see my response from a different account, too... :-)
> Kernel management is not an issue for me because recent SuSE 2.4.x
> kernels already include XFS support by default.
> What worries me is stability and tolerance to power failures and other
> "bad treatments". I have EXT2 here and I'm happy with it but since the
> servers would be located in client shops I wish to have something that
> doesn't need "human" input in such cases. Have you experienced (or
> heard) horror stories about XFS, expecially related to postgresql? Do
> you think JFS is better than XFS in this field too?
>
> Thanks again. Kind regards,
I _would_ recommend having a journalling filesystem as opposed to
ext2, from the perspective of atrocious fsck times, but I don't have a
metric that I am particularly confident in by which to evaluate JFS as
"better" than XFS, or vice-versa, from a "stability" perspective.
Neither has been available for long enough for there to be a large
body of results to report on.
I used to follow ReiserFS development (I was one of the "early
adoptors;" I have had filesystems on that FS since about 1998), and
heard [though did not personally experience] horror stories. I
recently had some minor data loss due to problems with ReiserFS, and
would definitely NOT recommend it for a PostgreSQL partition, as its
strengths don't fit with what PostgreSQL does. But I haven't been
following JFS/XFS mailing lists to hear horror stories.
--
output = ("cbbrowne" "@" "libertyrms.info")
<http://dev6.int.libertyrms.com/>
Christopher Browne
(416) 646 3304 x124 (land)
Christopher Browne wrote: > The best performance results I have seen on Linux systems have > involved the use of JFS. I found XFS to be a little slower, and it > has the distinct demerit that it is not in the 'official' kernel tree > yet, thereby meaning that you have to get into the pain of managing > heavily-patched kernels. The "kernel management" issue strikes me as > being a much bigger deal than the relatively minor performance > difference. Thank you for your answer (it's still me, now I'm using my "official" usenet account :)) Kernel management is not an issue for me because recent SuSE 2.4.x kernels already include XFS support by default. What worries me is stability and tolerance to power failures and other "bad treatments". I have EXT2 here and I'm happy with it but since the servers would be located in client shops I wish to have something that doesn't need "human" input in such cases. Have you experienced (or heard) horror stories about XFS, expecially related to postgresql? Do you think JFS is better than XFS in this field too? Thanks again. Kind regards, -- Cris Carampa (spamto:cris119@operamail.com) "Poveri fanatici comunisti, noglobal e affetti dalla sindrome anti-microsoft" (gli utenti Linux secondo un poster di ICOD)
On Fri, 2003-11-14 at 07:59, Cris Carampa wrote: > Christopher Browne wrote: > > > The best performance results I have seen on Linux systems have > > involved the use of JFS. I found XFS to be a little slower, and it > > has the distinct demerit that it is not in the 'official' kernel tree > > yet, thereby meaning that you have to get into the pain of managing > > heavily-patched kernels. The "kernel management" issue strikes me as > > being a much bigger deal than the relatively minor performance > > difference. > > Kernel management is not an issue for me because recent SuSE 2.4.x > kernels already include XFS support by default. > What worries me is stability and tolerance to power failures and other > "bad treatments". I have EXT2 here and I'm happy with it but since the > servers would be located in client shops I wish to have something that > doesn't need "human" input in such cases. Have you experienced (or > heard) horror stories about XFS, expecially related to postgresql? Do > you think JFS is better than XFS in this field too? We've used Linux XFS on RAID 0+1 here for some time. I don't see a big deal with kernel management either, as XFS is a pretty standard option these days. I've just received the okay to purchase maxed out SMP Opteron servers with Fiber Channel storage arrays to run Postgres, and those will be Linux with XFS file systems as well. JFS might be just as good or better than XFS performance-wise, but I've generally avoided it primarily because of the impression that it isn't as mature or thoroughly tested under Linux as XFS seems to be. And XFS seems to be generally pretty fast anyway. If you like XFS and are comfortable with it, I'd say use it. Cheers, -James Rogers jamesr@best.com
--- James Rogers <jamesr@best.com> wrote: > I've just received the okay to purchase maxed out SMP Opteron servers > with Fiber Channel storage arrays to run Postgres, and those will be > Linux with XFS file systems as well. Any particular brand of the storage array you'd recommend? Thanks __________________________________ Do you Yahoo!? Protect your identity with Yahoo! Mail AddressGuard http://antispam.yahoo.com/whatsnewfree
On Fri, 2003-11-14 at 10:54, ow wrote: > --- James Rogers <jamesr@best.com> wrote: > > I've just received the okay to purchase maxed out SMP Opteron servers > > with Fiber Channel storage arrays to run Postgres, and those will be > > Linux with XFS file systems as well. > > Any particular brand of the storage array you'd recommend? Thanks Not really. I haven't settled on any specific vendor at this point, I've just been given the okay to massively upgrade our core database systems to 64-bit mid-rangy iron, which I've scheduled for 1Q2004. Since we use Postgres on Linux, that means big Opteron boxes that are maxed out. If things keep going like they are, I could shortly be running on of the biggest PostgreSQL implementations out there. That could be both good AND bad. :-) Cheers, -James Rogers jamesr@best.com