[Gllug] AMD 64 X 2 / CentOS 4.3 x86-64 SMP RAID 5 slow

[Garry Heaton]

Nix wrote:

> I was wondering if perhaps it does like md does: if the first thing it 
> runs is a reintegration (i.e. to get the parity correct), and if it's not
>  as good as md is at deprioritizing the reintegration if real I/O comes 
> through, then you might see really slow I/O to start with.
> 
> (On non-degraded RAID-5 or RAID-6, you can normally expect things to be 
> *faster* than without RAID on the average, because you're getting your 
> data in parallel off two disks rather than one: it doubles the transfer 
> rate.)
> 

After rebuilding the array I noticed boot errors relating to not being able
to read the superblock on /dev/sda1 so looked at the 3rd disk added to the 2
that worked OK and decided it had to some kind of bad block problem. I then
used the 2 known good disks to setup a RAID 0 array then put the 3rd disk on
one of the motherboard's SATA connectors, loaded CentOS onto the RAID 0
array and looked at f'disk /dev/sdb'. There, sure enough were what appeared to
be dodgy entries from the previous efforts to get it working in a RAID 5
array. I wiped all the partitions using fdisk then used 'mkfs -t ext3' to
format /dev/sdb. All was good so no bad blocks according to 'e2fsck -f'.

I then rebuilt the RAID 5 array with all 3 disks and CentOS loaded with no
problem. None, that is, except the inexplicable slow first boot of all
applications when in runlevel 5. There has to be some kind of I/O bottleneck
somewhere.

Garry
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug