IDE performance drop between 2.4.23 and 2.6.0

From: Jean-Luc Fontaine
Date: Tue Dec 23 2003 - 07:45:02 EST


Using hdparm -Tt /dev/hdb:

on 2.4:
Timing buffer-cache reads: 168 MB in 2.01 seconds = 83.58 MB/sec
Timing buffered disk reads: 44 MB in 3.12 seconds = 14.10 MB/sec
on 2.6:
Timing buffer-cache reads: 172 MB in 2.02 seconds = 84.95 MB/sec
Timing buffered disk reads: 34 MB in 3.08 seconds = 11.04 MB/sec

Note the big drop of of 3 MB/sec on disk reads.

dmesg (edited) gives:

on 2.4:
VP_IDE: VIA vt82c586b (rev 41) IDE UDMA33 controller on pci00:07.1
hdb: IC35L040AVVN07-0, ATA DISK drive
hdb: 80418240 sectors (41174 MB) w/1863KiB Cache, CHS=79780/16/63, UDMA(33)
on 2.6:
VP_IDE: VIA vt82c586b (rev 41) IDE UDMA33 controller on pci0000:00:07.1
hdb: IC35L040AVVN07-0, ATA DISK drive
hdb: 80418240 sectors (41174 MB) w/1863KiB Cache, CHS=65535/16/63, UDMA(33)

hdparm -i output is identical between the 2 kernels and hdparm settings are identical as follows (except cylinders as above):

Model=IC35L040AVVN07-0, FwRev=VA2OAG0A, SerialNo=VNP210B2RAP0TB
Config={ HardSect NotMFM HdSw>15uSec Fixed DTR>10Mbs }
RawCHS=16383/16/63, TrkSize=0, SectSize=0, ECCbytes=52
BuffType=DualPortCache, BuffSize=1863kB, MaxMultSect=16, MultSect=16
CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=80418240
IORDY=on/off, tPIO={min:240,w/IORDY:120}, tDMA={min:120,rec:120}
PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 mdma2
UDMA modes: udma0 udma1 *udma2 udma3 udma4 udma5
AdvancedPM=yes: disabled (255) WriteCache=enabled
Drive conforms to: ATA/ATAPI-5 T13 1321D revision 1:

multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 79780/16/63, sectors = 80418240, start = 0

I noticed a difference in dmesg:
on 2.4:
Uniform Multi-Platform E-IDE driver Revision: 7.00beta4-2.4
on 2.6:
Uniform Multi-Platform E-IDE driver Revision: 7.00alpha2

Could that make a difference?

Please let me know if you'd like me to run more tests, patch kernels, ..., anything to regain the performance.


Many thanks in advance,

--
Jean-Luc Fontaine

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/