Disk performance problem
in [UK Linux]
|
From: Martin Liddle on 7 Apr 2010 17:02 I have two very similar systems. System one uses a Gigabyte GA-G31M-S2L motherboard; system two uses a Gigabyte GA-G31M-ES2L motherboard; RAM and processor are the same. Both systems use a Samsung HD322HJ SATA drive connected to the first SATA port. Both systems had a clean install of Centos 5.3 and have been upgraded to Centos 5.4. I have been aware that the disk performance of system two was not very good but had never got round to investigating. Today I used hdparm -t to measure the read performance. System one appears to be using SCSI emulation and hdparm -t /dev/sda returns a read performance of 115.8 MB/s which sounds respectable. System two appears to be using IDE mode and hdparm -t /dev/hda returns a read performance of 3.5MB/s which is obviously very poor (I assume it is using PIO mode). How do I improve the performance of system two? -- Martin Liddle, Tynemouth Computer Services, 3 Kentmere Way, Staveley, Chesterfield, Derbyshire, S43 3TW. Web site: <http://www.tynecomp.co.uk>.
From: Tom Anderson on 7 Apr 2010 18:03 On Wed, 7 Apr 2010, Martin Liddle wrote: > I have two very similar systems. System one uses a Gigabyte GA-G31M-S2L > motherboard; system two uses a Gigabyte GA-G31M-ES2L motherboard; RAM > and processor are the same. Both systems use a Samsung HD322HJ SATA > drive connected to the first SATA port. Both systems had a clean > install of Centos 5.3 and have been upgraded to Centos 5.4. I have been > aware that the disk performance of system two was not very good but had > never got round to investigating. Today I used hdparm -t to measure the > read performance. System one appears to be using SCSI emulation and > hdparm -t /dev/sda returns a read performance of 115.8 MB/s which sounds > respectable. System two appears to be using IDE mode and hdparm -t > /dev/hda returns a read performance of 3.5MB/s which is obviously very > poor (I assume it is using PIO mode). How do I improve the performance > of system two? Perhaps poke about in BIOS and kernel settings to see if DMA is turned on, and if not, whether it can be turned on. Sounds mad, but we had exactly this situation at work - brand new beefy workstations hobbling along. I think someone had to recompile the kernel to get them working properly in the end. I guess my answer is basically implicit in the question you're asking, though, and as such, is about as much use as a chocolate condom. Apologies. I'll see if i can get some more details out of the guy who did this. tom -- Gin makes a man mean; let's booze up and riot!
From: Doctor J. Frink on 8 Apr 2010 04:11 On 2010-04-07, Martin Liddle <news09(a)tynecomp.co.uk> wrote: > I have two very similar systems. System one uses a Gigabyte GA-G31M-S2L > motherboard; system two uses a Gigabyte GA-G31M-ES2L motherboard; RAM > and processor are the same. Both systems use a Samsung HD322HJ SATA > drive connected to the first SATA port. Both systems had a clean > install of Centos 5.3 and have been upgraded to Centos 5.4. I have been > aware that the disk performance of system two was not very good but had > never got round to investigating. Today I used hdparm -t to measure the > read performance. System one appears to be using SCSI emulation and > hdparm -t /dev/sda returns a read performance of 115.8 MB/s which sounds > respectable. System two appears to be using IDE mode and hdparm -t > /dev/hda returns a read performance of 3.5MB/s which is obviously very > poor (I assume it is using PIO mode). How do I improve the performance > of system two? It may be that the second system is configured in the BIOS to use a 'legacy' IDE mode. We have a supermicro based system that had this configured. The problem is that the driver doesn't quite work properly with the IDE emulation and DMA cannot be activated. Reconfiguring the BIOS to use native SATA or AHCI gives much better performance. I don't know of any way of forcing the IDE driver to work properly. Frink -- Doctor J. Frink : 'Rampant Ribald Ringtail' See his mind here : http://www.cmp.liv.ac.uk/frink/ Annoy his mind here : pjf at cmp dot liv dot ack dot ook "No sir, I didn't like it!" - Mr Horse
From: Martin Liddle on 8 Apr 2010 06:33 In message <slrnhrr3u6.k8l.frink(a)homer.cmp.liv.ac.uk>, Doctor J. Frink <frink(a)homer.cmp.liv.ac.uk> writes > >Reconfiguring the BIOS to use native SATA or AHCI gives much better >performance. I don't know of any way of forcing the IDE driver to work >properly. > Thanks. The relevant BIOS setting was "On-Chip SATA Mode" which was set to "Auto". I hadn't taken much notice as this is how it is set on the system that performs properly. However I have now changed it to "Enhanced" and the disk speed is increased by a factor of 40. -- Martin Liddle, Tynemouth Computer Services, 3 Kentmere Way, Staveley, Chesterfield, Derbyshire, S43 3TW. Web site: <http://www.tynecomp.co.uk>.
From: Tom Anderson on 8 Apr 2010 19:22
On Thu, 8 Apr 2010, Martin Liddle wrote: > In message <slrnhrr3u6.k8l.frink(a)homer.cmp.liv.ac.uk>, Doctor J. Frink > <frink(a)homer.cmp.liv.ac.uk> writes > >> Reconfiguring the BIOS to use native SATA or AHCI gives much better >> performance. I don't know of any way of forcing the IDE driver to work >> properly. > > Thanks. The relevant BIOS setting was "On-Chip SATA Mode" which was set > to "Auto". I hadn't taken much notice as this is how it is set on the > system that performs properly. However I have now changed it to > "Enhanced" and the disk speed is increased by a factor of 40. One of the scary things about modern hardware is that it can be underperforming by one and a half orders of magnitude, and you don't even notice for a while! tom -- In-jokes for out-casts |