RE: (Severe?) Performance issues with ext3 created on RAID5 with partman
Lennart Sorensen wrote:
> What speed does ext3 get on a single drive then?
That is NOT the point and I cannot measure it since all devices are
RAID5.
>> I should have stressed that reading from the ext3 is limited to 51
>> MB/s.
>
> Measured how? It isn't unusualy to have less speed from the
> filesystem than the raw device. There is some overhead after
> all, and you are seeking around. dd is sequential which is
> optimal for performance.
Now you could have read that from my mail :)
dd if=/some-large-file of=/dev/zero
reports 50 MB/s for all large ISO files I have tested it with (more than
5, I know what I am doing), and they are not fragmented, I have sorted
the fragmented ones out, they even do less speed.
dd if=/dev/md1 of=/dev/zero
reports 100 MB/s, tested with several seek offsets.
Now it should be very clear to you that the filesystem creates a 50%
overhead which is inacceptable for a 2800 MHz system. And please read
my mail again where it is explained that partman probably did not use
the "stride" option which might be _the_ reason for this overhead. Even
if you don't accept/understand my problem this still should be fixed as
the RAID5 HOWTO clearly states.
Please forward this mail to the bug-report mailinglist where you wrote to!
I only receive mails from there and cannot write to it and I don't want my
bug report to be disregarded because of misunderstandings!
My disks are full and I need to fix the problem with the fileserver
filesystem before it gets so full that I have no space for reformatting ;)
Thanks, Leo Bogert
Reply to: