Hi,
I made a HW patch.
Added a Seagate U10 ST320423A (20,4 GB)
<6>Uniform Multi-Platform E-IDE driver Revision: 6.31
<4>ide: Assuming 30MHz system bus speed for PIO modes
<4>PIIX3: IDE controller on PCI bus 00 dev 39
<4>PIIX3: chipset revision 0
<4>PIIX3: not 100% native mode: will probe irqs later
<4> ide0: BM-DMA at 0xffa0-0xffa7, BIOS settings: hda:pio, hdb:pio
<4> ide1: BM-DMA at 0xffa8-0xffaf, BIOS settings: hdc:pio, hdd:pio
<4>hda: WDC AC33200L, ATA DISK drive
<4>hdb: ST320423A, ATA DISK drive
<4>hdc: WDC AC22100H, ATA DISK drive
<4>ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
<4>ide1 at 0x170-0x177,0x376 on irq 15
<6>hda: 6346368 sectors (3249 MB) w/256KiB Cache, CHS=787/128/63, (U)DMA
<6>hdb: 40011300 sectors (20486 MB) w/512KiB Cache, CHS=2490/255/63,
(U)DMA
<6>hdc: 4124736 sectors (2112 MB) w/128KiB Cache, CHS=4092/16/63, DMA
And did my streaming test against it instead of the WDCAC33200L.
Performance doubled as it should, sustained throuput should be a lot
better for the new Seagate.
But more interesting is that latencies went way down - why ?
(running with Andrews low latency patch)
# first run only DMA, readahead 8
Jul 15 01:19:51 dox kernel: Latency 5ms PID 2 % kswapd
Jul 15 01:26:24 dox kernel: Latency 7ms PID 0 % swapper
# second run, multicount 8, 32bit, unmaskirq, DMA, readahead 8
Jul 15 01:33:47 dox kernel: Latency 8ms PID 2 % kswapd
Jul 15 01:33:47 dox kernel: Trace: [shrink_mmap+66/532]
Jul 15 01:34:04 dox kernel: Latency 5ms PID 328 % bash
Jul 15 01:34:04 dox kernel: Trace: [__mark_buffer_dirty+54/56]
Jul 15 01:34:09 dox kernel: Latency 11ms PID 2 % kswapd
Jul 15 01:34:09 dox kernel: Trace: [shrink_mmap+66/532]
Jul 15 01:34:36 dox kernel: Latency 6ms PID 2 % kswapd
Jul 15 01:34:37 dox kernel: Trace: [shrink_mmap+66/532]
Jul 15 01:35:54 dox kernel: Latency 5ms PID 2 % kswapd
Jul 15 01:35:55 dox kernel: Trace: [shrink_mmap+66/532]
Jul 15 01:36:02 dox kernel: Latency 19ms PID 0 % swapper
Jul 15 01:43:46 dox kernel: Latency 7ms PID 0 % swapper
That is all above 5 ms two runs, compare that to the old one.
(a run = 3 * write 150 MB, 3 * copy 150 MB,
3 * read 300 MB, 3 * diff 150 MB)
kernel folks:
* Why do I get better latencies?
* Why does latencies get worse when enabling good options?
/RogerL
--
Home page:
http://www.norran.net/nra02596/
