Slow disk makes machine unusable

designator · 12-24-2007, 08:50 AM

I have a couple of old 120GB IDE Western Digital disks in my computer connected to a controller with JMicron 363 chipset. Problem is, they are insanely slow. They are formatted with ext3. 1% reserved.

Code:

hdparam -t /dev/hdc
/dev/hdc:
 Timing buffered disk reads:    4 MB in  3.64 seconds =   1.10 MB/sec

This wouldn't really be a problem, but any sort of io access to either disk brings my computer to a crawl. Reading slows it down to basically unusable, and writing just kills it.

I usually connect to this machine remotely and any attempt to write to either drive kills all ssh sessions (along with any other network services like NAT,) and prevents me from reconnecting for about 15-20 mins.

When I am finally able to reconnect there is nothing in the logs that shows any sort of errors.

These disks are old, but I expect them to perform at least a little better than cd-roms :-) It would also be really nice if there was a way to background rw access so that it doesn't dominate the computer.

Any ideas welcome! I'd love to hear that it is just some dumb setting I need to change and I'll get 50MB/s out of them :-)

Happy holidays.

Lenard · 12-24-2007, 09:25 AM

Is DMA enabled for these drives??? hdparmd /dev/hdc

See: http://www.linuxdevcenter.com/pub/a/...29/hdparm.html

farslayer · 12-24-2007, 10:11 PM

Are you sure the issue isn't with the controller and not the drives ? The Jmicron controller has been a royal pain in the past..

https://bugs.launchpad.net/ubuntu/+s....15/+bug/57502

http://forums.gentoo.org/viewtopic-t...ighlight-.html

I've installed Linux on drives older than the 120GB drives you mentioned and never had a perfomance issue as severe as you describe.

designator · 12-26-2007, 02:38 AM

Well, I did recently move the drives from HighPoint RocketRAID controller, but they have been causing similar problems before, I just never bothered to test the transfer rate because I used them for archiving old stuff I didn't really care for - so not much access. Also, I had 3 of these in RAID 0, so they had to work at least a little faster. Now I want to compress the data on them so basically read whole drive, the write compressed data back to the entire drive.

Drives are not in RAID right now.

Unfortunately, I'm about 150 miles away from this machine so I don't have the convenience of rebooting in single user mode if I mess up hdparam settings, but I'll definitely play around with those next time I can log in locally. Thanks for the link!

DMA is off:

Code:

/sbin/hdparm /dev/hdc

/dev/hdc:
 multcount    = 16 (on)
 IO_support   =  0 (default 16-bit)
 unmaskirq    =  0 (off)
 using_dma    =  0 (off)
 keepsettings =  0 (off)
 readonly     =  0 (off)
 readahead    = 256 (on)
 geometry     = 16383/255/63, sectors = 234441648, start = 0

designator · 12-30-2007, 12:28 PM

Quote:

Originally Posted by Lenard

Is DMA enabled for these drives??? hdparmd /dev/hdc

See: http://www.linuxdevcenter.com/pub/a/...29/hdparm.html

Thanks again for that link. My IO-Wait is still around 70%, but transfer rate improved greatly and now that it doesn't lock the interrupts -- it doesn't kill network connections.

Code:

/sbin/hdparm -t /dev/hdd

/dev/hdd:
 Timing buffered disk reads:  150 MB in  3.01 seconds =  49.91 MB/sec