Western Digital Velociraptor WD740ADFD-00NLR5 - NCQ performance falls off a cliff wi queue depth greater than two

[Tim Small <tim@seoss.co.uk>]

Hello,

I have what looks like a firmware or design performance bug with some 
74G WD Velociraptors WD740ADFD-00NLR5 (firmware 21.07QR5).  Sequential 
read speed (as measured with 'hdparm -t') drops-off with NCQ depths 
greater than 2.

Queue-depth(cmds) Speed(MB/s)

1   84
2   84
3   46
5   39
10   39
20   44
31   36

The speed given is the best of 10 - the system was under light load (I 
couldn't completely unload it, unfortunately).

Western Digital helpfully suggested that I contact my Linux vendor...


The system is a Tyan server with a Core2Quad Q6600  @ 2.40GHz, and an 
ICH9 running Debian 5.0 - kernel 2.6.26-2-openvz-amd64, SATA ports in 
non-RAID/AHCI/native mode.

I measured the speed like this:

for i in {1..31} ; do echo $i > /sys/block/sda/device/queue_depth ; echo 
-n "$( cat /sys/block/sda/device/queue_depth | head -c -1 )," ; for j in 
{1..10} ; do (hdparm -t /dev/sda | egrep -o 
'[[:digit:]]+\.[[:digit:]]+[[:space:]]+MB/sec' | head -c -1 ) ; done ; 
echo ; done | sed 's/ MB\/sec/,/g'

... and created a chart and max speed data in openoffice:

http://buttersideup.com/files/wd-raptor-NCQ-bug-1.pdf

with a second run here:

http://buttersideup.com/files/wd-raptor-NCQ-bug-2.pdf

Any comments or suggestions welcome - I'd be happy to repeat the test 
using a different benchmark if anyone can suggest a command-line they'd 
like run?

Cheers,

Tim.

-- 
South East Open Source Solutions Limited
Registered in England and Wales with company number 06134732.  
Registered Office: 2 Powell Gardens, Redhill, Surrey, RH1 1TQ
VAT number: 900 6633 53  http://seoss.co.uk/ +44-(0)1273-808309