Hi all,
I've been doing some bonnie benchmarks with the same hard drives on the
same machine under a few kernels. I found that a single IDE drive can
be very fast once tuned with hdparm (23 MBps block I/O). But when two
drives on different IDE channels are both used at the same time some
major bottleneck is hit... performance of *both* drives together is
just slightly better than one drive by itself.
- Has anyone else has seen similar problems?
- What could be limiting the two disk drive performance so much?
(hardware controller? IDE disk driver? kernel I/O system?)
- Suggestions to improve performance of multi-disk I/O?
Quintela's VM patches work wonderfully here preventing "VM: killing
process X", system lockups, and allow bonnie tests to actually run to
completion (instead of erroring out which started in -pre7). The system
is slightly hard to use when heavy disk I/O is happening, but *much*
better than recent 2.3.* kernels.
Since 2.3.99-pre* the VM issues aren't solved yet I'm mainly pointing
this out as an observation so someone can look into heavy disk I/O
performance in 2.3.* after the VM problems are fixed.
- Disk _output_ appears to have been affected by the first patch set...
- With the newest Quintela patch set _output_ is slowed more and block
_input_ is also being affected.
This poor performance _could_ be caused by addition swap activity under
the patched kernels, but I do not have vmstat runs to match all of the
bonnie tests to verify the swap activity is new and is enough to be
responsible for the single disk performance losses.
Later,
Mike Simons
Both drives were initialized at boot via "hdparm -m 16 -u 1 -c 1 -d 1"
Bonnie++ was found at http://www.coker.com.au/bonnie++/
all tests used "-d . -s 256 -n 0" as the options to bonnie.
"-p 2" and "-y" were used to start the multi-disk tests.
Quintela's patches as I have applied them are available
(for a limited time only ;) at http://moria.simons-clan.com/~msimons/
[note: these numbers are not the true average of several runs...
they are just the output of a sample run. the actual averages should
appear to be within 10% of the numbers shown below which is good enough
to get the point across.]
Version 1.00a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
MB K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
2.2.14 ===============================================================
hda 256 7774 97 24267 21 10049 21 8069 95 23229 21 47.5 0
hdc 256 7866 98 22830 20 5702 10 7194 95 18741 15 40.0 0
hda *and* 256 3791 47 11090 9 5684 12 3723 49 13454 11 25.8 0
hdc *and* 256 3869 48 11179 9 5505 12 3705 49 13187 12 25.6 0
===========================================================================
2.3.99-pre9-pre1 + quintela patch set 1 ===================================
hda 256 7621 96 14375 13 8091 12 6616 78 21686 21 42.2 0
hdc 256 7690 97 18216 16 7556 11 6398 84 22133 21 41.1 0
hda *and* 256 3194 40 9984 8 5393 8 3801 50 13266 14 23.5 0
hdc *and* 256 3467 43 9832 8 5381 8 3769 50 13842 14 23.9 0
===========================================================================
2.3.99-pre9-pre3 + quintela patch set 2 ===================================
hda 256 7140 90 16162 14 8179 12 7090 83 20079 19 43.5 0
hdc 256 7432 93 16418 14 6607 10 6646 78 14330 13 39.2 0
hda *and* 256 3540 44 6581 5 5167 7 4025 47 11184 11 24.6 0
hdc *and* 256 3192 40 6582 5 5051 8 4003 47 10440 11 23.4 0
===========================================================================
Machine specs:
Pentium III 500 Mhz, 128 Megs RAM
from kernel config:
CONFIG_BLK_DEV_IDEPCI=y
CONFIG_IDEPCI_SHARE_IRQ=y
CONFIG_BLK_DEV_IDEDMA_PCI=y
CONFIG_IDEDMA_PCI_AUTO=y
CONFIG_BLK_DEV_IDEDMA=y
CONFIG_IDEDMA_PCI_EXPERIMENTAL=y
CONFIG_BLK_DEV_PIIX=y
CONFIG_PIIX_TUNING=y
/proc/interrupts
14: 100689 XT-PIC ide0
15: 115092 XT-PIC ide1
/proc/ioports
ffa0-ffaf : Intel Corporation 82371AB PIIX4 IDE
ffa0-ffa7 : ide0
ffa8-ffaf : ide1
/proc/pci
Bus 0, device 0, function 0:
Host bridge: Intel Corporation 440BX/ZX - 82443BX/ZX Host bridge (rev 3).
Master Capable. Latency=64.
Prefetchable 32 bit memory at 0xf0000000 [0xf3ffffff].
Bus 0, device 7, function 1:
IDE interface: Intel Corporation 82371AB PIIX4 IDE (rev 1).
Master Capable. Latency=32.
I/O at 0xffa0 [0xffaf].
lspci -vvv:
00:00.0 Host bridge: Intel Corporation 440BX/ZX - 82443BX/ZX Host bridge \
(rev 03)
Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- \
ParErr- Stepping- SERR+ FastB2B-
Status: Cap+ 66Mhz- UDF- FastB2B- ParErr- DEVSEL=medium \
>TAbort- <TAbort- <MAbort+ >SERR- <PERR-
Latency: 64 set
Region 0: Memory at f0000000 (32-bit, prefetchable) [size=64M]
Capabilities: [a0] AGP version 1.0
Status: RQ=31 SBA+ 64bit- FW- Rate=21
Command: RQ=0 SBA- AGP- 64bit- FW- Rate=
00:07.1 IDE interface: Intel Corporation 82371AB PIIX4 IDE (rev 01) \
(prog-if 80 [Master])
Control: I/O+ Mem- BusMaster+ SpecCycle- MemWINV- VGASnoop- \
ParErr- Stepping- SERR- FastB2B-
Status: Cap- 66Mhz- UDF- FastB2B+ ParErr- DEVSEL=medium \
>TAbort- <TAbort- <MAbort- >SERR- <PERR-
Latency: 32 set
Region 4: I/O ports at ffa0 [size=16]
from dmesg:
Uniform Multi-Platform E-IDE driver Revision: 6.30
ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx
PIIX4: IDE controller on PCI bus 00 dev 39
PIIX4: chipset revision 1
PIIX4: not 100% native mode: will probe irqs later
ide0: BM-DMA at 0xffa0-0xffa7, BIOS settings: hda:DMA, hdb:pio
ide1: BM-DMA at 0xffa8-0xffaf, BIOS settings: hdc:DMA, hdd:DMA
hda: WDC WD102BA, ATA DISK drive (hdparm -i -> FwRev=16.13M16)
hdc: WDC WD43AA, ATA DISK drive (hdparm -i -> FwRev=29.05T29)
hdd: CRD-8400B, ATAPI CDROM drive
hda: 20028960 sectors (10255 MB) w/2048KiB Cache, CHS=1246/255/63, UDMA(33)
hdc: 8421840 sectors (4312 MB) w/2048KiB Cache, CHS=8355/16/63, UDMA(33)
hdd: ATAPI 48X CD-ROM drive, 128kB Cache, UDMA(33)
===========
fizban:/proc/ide# cat piix
Intel PIIX4 Ultra 33 Chipset.
--------------- Primary Channel ---------------- Secondary Channel -------------
enabled enabled
--------------- drive0 --------- drive1 -------- drive0 ---------- drive1 ------
DMA enabled: yes no yes yes
UDMA enabled: yes no yes yes
UDMA enabled: 2 X 2 2
UDMA
DMA
PIO
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue May 23 2000 - 21:00:23 EST