Linux Software RAID-5 trouble/degraded performance

[Joffer]

I've got a Pentium III 667MHz 256MB RAM and 3 WD800 (80GB Western Digital disks hda/b/d) with a CD-ROM as hdc. I installed TSL 2.1 (Trustix Secure Linux) on this box and configured it with a single /boot partition (hda 128MB), two swap (hdb&hdd 128MB), three 128MB sw-raid partitions for /tmp and three large sw-raid partitions for / (root). Both sw-raid partitions were set up as Raid-5.

Now to my trouble - hdd is running at degraded speed, so system is running with a degraded performance. Running hdparm -Tt /dev/hdd gave me a "Timing buffered disk reads" of only 3MB/s on hdd, while over 30MB/s on hda and hdb.

I've included some config files and outputs from tests below. After this, I installed TSL 2.1 again without Raid-5 and just mountet hdb and hdd as /var and /tmp. hdparm testing on hdd gave me full 30MB/s in the "Timing buffered disk reads" test... I got this raid-5 results from a TSL 2.0 install last week also.

Any suggestions / solutions to why I get so degraded hdd speed and thus bad system responses?

/Christopher


Config files and outputs:

root@c3po ~# cat /etc/fstab
/dev/md0 / reiserfs defaults 1 1
/dev/hda1 /boot ext3 defaults 1 2
none /dev/pts devpts gid=5,mode=620 0 0
none /proc proc defaults 0 0
/dev/md1 /tmp reiserfs defaults 1 2
/dev/hdb1 swap swap defaults 0 0
/dev/hdd1 swap swap defaults 0 0
/dev/cdrom /mnt/cdrom iso9660 noauto,owner,kudzu,ro
0 0

root@c3po ~# cat /etc/mdadm.conf
DEVICE /dev/hda3 /dev/hdb3 /dev/hdd3
ARRAY /dev/md0 level=raid5 devices=/dev/hda3,/dev/hdb3,/dev/hdd3
DEVICE /dev/hda2 /dev/hdb2 /dev/hdd2
ARRAY /dev/md1 level=raid5 devices=/dev/hda2,/dev/hdb2,/dev/hdd2

root@c3po ~# hdparm /dev/hd{a,b,d}

/dev/hda:
multcount = 16 (on)
IO_support = 0 (default 16-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 9729/255/63, sectors = 156301488, start = 0

/dev/hdb:
multcount = 16 (on)
IO_support = 0 (default 16-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 9729/255/63, sectors = 156301488, start = 0

/dev/hdd:
multcount = 16 (on)
IO_support = 0 (default 16-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 9729/255/63, sectors = 156301488, start = 0

root@c3po ~# hdparm -tT /dev/hd{a,b,d}

/dev/hda:
Timing buffer-cache reads: 96 MB in 2.01 seconds = 47.76 MB/sec
Timing buffered disk reads: 92 MB in 3.05 seconds = 30.16 MB/sec

/dev/hdb:
Timing buffer-cache reads: 116 MB in 2.06 seconds = 56.31 MB/sec
Timing buffered disk reads: 92 MB in 3.00 seconds = 30.67 MB/sec

/dev/hdd:
Timing buffer-cache reads: 116 MB in 2.03 seconds = 57.14 MB/sec
Timing buffered disk reads: 10 MB in 3.21 seconds = 3.12 MB/sec

 

[drag]

Well, On IDE stuff the Master sets the speed.

So I am going to speculate that the cdrom drive being master is screwing up your performance. Try making the harddrive the master, and the cdrom the slave.


Plus usually when setting DMA access most modern drives automaticly set the fastest IDE/ATA transfer modes, however this isn't always the case.


To see what modes they are using check out the output of hdparm -i /dev/hd{a,b,c,d}|less.

You can set them with the -X option, check the man file for details.

(just a shot in the dark, don't know much about raid)

[edit]
I found this webpage here

So I suppose the CDROM master vs slave thing is only a issue on IDE controllers that can't set independant data transfer types....

 

[Abzstrak]

kinda odd, I would make sure the cdrom is on its own controller too, just add a $15 card to the system... Also, the cable could be bad which cause alot of speed problems, or the drive could be dying.

 

[13black]

I had a similar problem running raid 0. System would degrade to ata-100, from ata-133. In my case it was my rounded IDE cables, they were too long. When I went back to 24 inch cables I got my ata-133 speeds.