Crappy SATA performance on asus p5q

[curaga]

Delete all data: boot from a livecd, cat /dev/urandom > /dev/sda

That only writes one pass with pseudo-random data, if you'd like more, DBAN has a good selection.

[Dan Harris]

Yes, 100% sure. The 750GB model is based on 3 250GB pattern where the 640GB is based on 2 320GB pattern. The size per patter might be a little higher and not deactivated to serve as "backup space" for those areas that tend to get damaged over time, so in fact the patterns might be 340GB or something like this.

What stays is that the 320, 640 and 1000GB drives are built upon the same patters. So they are more dense and the header has to move less while reading (and moving the read head takes a huge potion of time needed for reading/writing).

Well I have just changed the drive to the 1TB and this is from hdparm

/dev/sda:
Timing cached reads: 5132 MB in 2.00 seconds = 2566.77 MB/sec
Timing buffered disk reads: 274 MB in 3.01 seconds = 90.91 MB/sec

Still a little bit short of what you are getting. and not that much better than the 750. Any ideas.

Dan

[rbmorse]

Looks pretty nominal to me. What do you think you should be getting?

[Dan Harris]

Looks pretty nominal to me. What do you think you should be getting?

Well if you read the previous message ivanovic is getting 107 MB/Sec from his samsung sata 300 disk.

That would be nice.

Dan

[StringCheesian]

So with the exact same model of HD (correct?) ivanovic is getting 107MB/s and Dan Harris is getting 91MB/s.

What motherboards, kernel versions, and IO schedulers are you two using?

[Dan Harris]

So with the exact same model of HD (correct?) ivanovic is getting 107MB/s and Dan Harris is getting 91MB/s.

What motherboards, kernel versions, and IO schedulers are you two using?

I am using debian lenny 2.6.26-amd64 motherboard asus p5q as for scheduler I think I am using the following.

[ 1.041322] io scheduler cfq registered (default)

My HDD SAMSUNG HD103UJ so yes both same

Dan

[Dan Harris]

I am using debian lenny 2.6.26-amd64 motherboard asus p5q as for scheduler I think I am using the following.

[ 1.041322] io scheduler cfq registered (default)

My HDD SAMSUNG HD103UJ so yes both same

Dan

After about 6 hours of use last night the HD performance got progressively better and this morning hdparm show the following

/dev/sda:
Timing cached reads: 4942 MB in 2.00 seconds = 2471.38 MB/sec
Timing buffered disk reads: 326 MB in 3.00 seconds = 108.58 MB/sec

so all is well now. I guess a new hdd need running in for several hours to get it up to its best performance. I must admit when I was getting results as low as 78 MB/Sec I thought that I had a dud but a few hours later and all it well.

Thanks for all your help and info.

Dan

[hdas]

Although not directly related, here is something I experienced. I have a Gigabyte EP43-DS3L mobo with ICH10. And using a a seagate sata2 500gb hard disk - ST3500320AS. There is something funny about the AHCI controller. With kernels 2.6.26 and older, with AHCI turned on, I used to get 55MB/s and with AHCI off, I get 110MB/s. (Note that the SATA controller is enabled in both cases in the bios.) However, in Windows, it was 110MB/s in both cases. With latest kernel 2.6.27, it was fixed and now I get close to 110MB/s with AHCI enabled too. (The only thing I hate about enabling AHCI is that some crappy intel ahci bios application called iSrc takes more time at starting up, than an entire os bootup possibly.)

The only difference I see between 2.6.26 and 2.6.27 is that in dmesg output, I now see a new extension called 'ems' :

ahci 0000:00:1f.2: AHCI 0001.0200 32 slots 6 ports 3 Gbps 0x3f impl SATA mode
ahci 0000:00:1f.2: flags: 64bit ncq sntf stag pm led clo pmp pio slum part ems

In general, I haven't seen any benefit of this AHCI regarding performance on any machine - including my notebooks with ich7m and ich8m. May be just some power saving or such.