You actually expect an IDE-drive to be faster than that?

Hi,

how fast should your drives be?

Have you benchmarked it under another linux-kernel? What are the results?

Here are my results for a rather old computer:
(Pentibumm 700, 256 MB RAM, some kind of WDC drives)

/dev/hda:

 Timing buffer-cache reads:   392 MB in  2.00 seconds = 196.00 MB/sec

 Timing buffered disk reads:   80 MB in  3.02 seconds =  26.49 MB/sec

In my experience the existence of a windows partition at the start of the hard drive leads to slower results for buffered disk reads. I once posted this on the kernel list (archived in this thread) because it puzzled me, but seemingly nobody there didn't really know either - or nobody cared enough. Mostly everybody just dismissed it as "can't be", the only reply being backed by some knowledge was Andre Hedricks, but he couldn't explain it either.

Anyway, if there is a windows partition, retest with the first linux partition on that disk. I guess you'll get better results.

my machine : XP1800+/KT3V/512M/SEAGATE80G(2M)

I set harddisk on DMA5.

#/sbin/hdparm -tT /dev/hda
/dev/hda:
Timing buffer-cache reads: 128 MB in 0.14 seconds =295.24 MB/sec
Timing buffered disk reads: 64 MB in 2.73 seconds = 53.42 MB/sec

#/sbin/hdparm -tT /dev/hda
/dev/hda:
Timing buffer-cache reads: 128 MB in 0.14 seconds =296.76 MB/sec
Timing buffered disk reads: 64 MB in 2.73 seconds = 56.30 MB/sec

Another one?
This is a 1 year old laptop hard disk running kerel 2.6.1. I played around with all hdparm settings quite a lot, but can't seem to improve my performance.

# hdparm -tT /dev/hda2

/dev/hda2:
Timing buffer-cache reads: 1224 MB in 2.00 seconds = 611.18 MB/sec
Timing buffered disk reads: 26 MB in 3.04 seconds = 8.55 MB/sec

And there is a very strong variance, too: I've seen this go down to a mere !!! 3.5 MB/sec !!!
Sometimes, starting firefox takes more than half a minute on this drive.

Hi !

I also notified lower IDE performance with kernel 2.6.0 comparing to kernel 2.4.0. On my system I have two disks configured to RAID-0 (stripping). I have got VIA chipset. Results of hdparm -t on kernel 2.4.24 differs on kernel 2.6.0 like the following:

On kernel 2.6.0:

hdparm -t /dev/hda - gives about 40 MB/s
hdparm -t /dev/md7 - gives about 38 MB/s (even lower !)

On kernel 2.4.24:

hdparm -t /dev/hda - about 40 MB/s
hdparm -t /dev/md7 - about 80 MB/s (this is what I wanted !)

Do you have solution of the performace problem on 2.6.x kernel ?

Regards

I just recently put in a new disk im my old,old system (PII-450MHz, 384Mb), a WD80Gb/8Mb. Running 2.4.22 (Gentoo, kernel 2.4.22-r4) I can consistently get about 25Mb/sec in hdparm/buffered disk reads. I tried out 2.6.3 (Gentoo 2.6.3-r3 with gentoo patches), and I can get no more than 12Mb/sec under that kernel (sometimes even *much* lower). And yes, I have carefully checked that DMA-transfers are enabled, that the kernel is configured with all relevants settings (including use of IDE-DMA, chip set (Intel BX440) and what have you)).

Strange ...

My ICH5 is doing great, but a friend of mine with NFORCE2 gets 15MB/sec less than 2.4 with 2.6....

my results:
#uname -a
Linux garfield 2.6.3 #4 SMP Sat Feb 21 00:36:32 EET 2004 i686 Intel(R) Pentium(R) 4 CPU 2.60GHz GenuineIntel GNU/Linux

#hdparm -tT /dev/hda
/dev/hda:
Timing buffer-cache reads: 3484 MB in 2.00 seconds = 1741.39 MB/sec
Timing buffered disk reads: 160 MB in 3.03 seconds = 52.85 MB/sec

I have done two more tests with 2.5.52 and 2.5.64:

On 2.4.20: hdparm -t /dev/hda - gives about 19.80 MB/s
On 2.5.30: hdparm -t /dev/hda - gives about 19.80 MB/s
On 2.5.52: hdparm -t /dev/hda - gives about 19.80 MB/s
On 2.5.64: hdparm -t /dev/hda - gives about 15.70 MB/s
On 2.5.75: hdparm -t /dev/hda - gives about 15.70 MB/s
On 2.6.3: hdparm -t /dev/hda - gives about 15.70 MB/s

The next try... (rolling of a drum) 2.5.58!!! :D

I have tested the IDE performance of two more kernels on my computer, in order to find where was the critical change in the driver.
The command used to do the tests was # hdparm -t /dev/hda, and I've run the command at less 5 times in each test
The same kernel can throw different results on differents bootings, but I take the best of all.

Without DMA (# hdparm -d0 /dev/hda)
2.4.20: 5.70 MB/s in all bootings (aprox)
2.5.30: ---- hdparm hangs the disk :(
2.5.52: 5.67, 3.63 MB/s in 2 different bootings
2.5.58: 5.66 MB/s
2.5.61: 5.66 MB/s
2.5.64: 5.68 MB/s
2.5.75: 5.68 MB/s
2.6.3: 5.65 MB/s

With DMA (# hdparm -d1 /dev/hda)
This is the interesting
2.4.20: 19.88 MB/s
2.5.30: 20.00, 19.00 MB/s
2.5.52: 19.85 MB/s
2.5.58: 19.70 in one test, 19.00 in three tests
2.5.61: 15.37 MB/s here begins low performance
2.5.64: 15.73 MB/s
2.5.75: 15.69 MB/s
2.6.3: 15.80, 15.70, 15.37 MB/s

I think that the critical change is in 2.5.59, 60 or 61, so I'm downloading now 2.5.59

Why is "buffer-cache reads" less when X is running?

/dev/sda:
Timing buffer-cache reads: 3908 MB in 2.00 seconds = 1954.30 MB/sec
Timing buffered disk reads: 168 MB in 3.03 seconds = 55.38 MB/sec

This is a single Seagate 80GB SATA drive on the ICH5. Using kernel 2.6.4 with Con Koliva's patch set. Several things running at same time. I ran this test before I installed much else, and with various other patch sets such as mm and the stock kernel, and had similar results with all. Only a few MB/sec higher when I didn't have so much running. I use the Intel SATA driver in the SCSI section. These results are pretty standard from what I have seen. Would be nice to be able to use hdparm to tweak some settings, but from what I know of drive performance, this is pretty close to the real world limit of the drive anyway.

2x Maxtor 80GB 8MB cache. hda is on integrated nforce2, hdk is on promise pci controler. md0 is raid1, md4 is raid0. Barton 2500+, 333 bus, 1Gb ram, nforce2 board w/ integrated video running X. kernel 2.6.4-genoo-r1.

alumiguin root # hdparm -tT /dev/hda

/dev/hda:
Timing buffer-cache reads: 1356 MB in 2.00 seconds = 677.43 MB/sec
Timing buffered disk reads: 168 MB in 3.00 seconds = 55.92 MB/sec
alumiguin root # hdparm -tT /dev/hdk

/dev/hdk:
Timing buffer-cache reads: 1320 MB in 2.00 seconds = 659.77 MB/sec
Timing buffered disk reads: 168 MB in 3.01 seconds = 55.80 MB/sec
alumiguin root # hdparm -tT /dev/md0

/dev/md0:
Timing buffer-cache reads: 1404 MB in 2.00 seconds = 701.76 MB/sec
Timing buffered disk reads: 168 MB in 3.00 seconds = 55.93 MB/sec
alumiguin root # hdparm -tT /dev/md4

/dev/md4:
Timing buffer-cache reads: 1292 MB in 2.00 seconds = 645.45 MB/sec
Timing buffered disk reads: 286 MB in 3.00 seconds = 95.25 MB/sec

I get the same thing as you except when i go back into 16-bit mode - i get:

/dev/hdg:
Timing buffer-cache reads: 880 MB in 2.00 seconds = 439.19 MB/sec
Timing buffered disk reads: 170 MB in 3.03 seconds = 56.06 MB/sec

When i'm in 16-bit mode. Strange eh?

I have an ICH5. I was getting about 30MB/sec transfer.

Then I noticed that my disk was configured as a slave even though I only have one drive attached to the controller. I popped open the case, changed the cables around and now I get 48MB/sec.

My SATA drive does 55MB/sec.

Dell PE400SC, 2.8Ghz