xfs
[Top] [All Lists]

Re: Incredibly slow rebuild of small access_db on xfs partition

To: Jarek Luberek <jarek@xxxxxxxxxx>
Subject: Re: Incredibly slow rebuild of small access_db on xfs partition
From: Russell Cattelan <cattelan@xxxxxxxxxxx>
Date: Sat, 05 May 2001 17:33:51 -0500
Cc: linux-xfs@xxxxxxxxxxx, Keith Owens <kaos@xxxxxxxxxxxxxxxxx>
References: <17466.989071943@ocs3.ocs-net> <01050522144000.00690@marvin> <01050522293200.00688@marvin>
Sender: owner-linux-xfs@xxxxxxxxxxx
Jarek Luberek wrote:

This is probably the same problem as rpm --rebuilddb

the db package apparently query the files system for the
"optimal" transfer size, on XFS this is 64k on ext2 it's 4k.

The theory is that db is using a 64k blocks for every record rather
than 4k?

You can turn this number down in xfs... hmm maybe I can make it
a tunable parameter, that way we can test this theory quickly.



> Hi,
>
> I just went back to the xfs partition an it seems that
> every time I run the makemap command I get a file
> of the size  154599424. The correct size is 28672.
>
> Finally, the output from strace -c on the xfs partition:
>
> % time     seconds  usecs/call     calls    errors syscall
> ------ ----------- ----------- --------- --------- ----------------
>  53.34    4.030099         788      5113           write
>  44.63    3.371511         651      5182           read
>   1.87    0.141185          14     10284           lseek
>
> The time of the makemap command on the xfs partition
> was this time
>
> real    0m24.263s
> user    0m0.070s
> sys     0m7.650s
>
> Greetings,
> jarek

--
Russell Cattelan
cattelan@xxxxxxxxxxx




<Prev in Thread] Current Thread [Next in Thread>