Yesterday i was creating an ide raid0 array using
kernel 2.2.13 with
raid0145-19990824-2.2.11 patch (for raid v0.90)
ide.2.2.13.19991104 patch (for my hpt 366 controller)
on
AMD K6-2-266
128MB SDRAM
hda = Quantum KA 18GB (bonnie=20MB/s)
hdc = Quantum EX 6.4GB (bonnie=14MB/s)
hde = Quantum CR 6.4GB (~16MB/s i think)
hdg = Quantum CR 6.4GB ( ditto )
Onboard via chipset (Apollo MVP3) controlling hda + hdc
PCI HPT366 (rev 1) IDE card (Abit Hotrod) controlling hde + hdg
Other hardware Adaptec 2940 cotnrolling 2 cdroms, realtek 8139, riva tnt
Anyway, im benchmarking away using bonnie, it was going great guns using a 2
way raid0, getting linear performance increase, getting good speed, tried
various combinations.
1) 20MB/s on hda + hdc
2) 29MB/s on hde + hdg,
3) 25MB/s on hda + hde,
4) 25MB/s on hdc + hdg,
I would have thought 3 + 4 would be different as hdc is faster than hda,
anyway, point is i could use any combiantion of two drives in a raid0
without a timeout problem.
Next i was going to some benchmarks with a 3way raid0, but i kept getting
those damn pesky DMA and IRQ timouts, like
hde: timeout waiting for DMA
hde: irq timeout: status=0x58 { DriveReasy SeekComplete DataRequest }
hde: DMA disabled
also read_intr: status=0x50 { DriveReady SeekComplete }
I only get them on drive hde and hdg (the hpt366 controller)
I got bonnie to run without timeouts on hda hdc hde at 25MB/s,
When i tried hde+hdg+ either hda or hdc i consistently get timeouts.
I tried compiling the kernel with the hpt366 fast interrupt option but i
didnt see any difference (and wasnt really sure what it was supposed to do,
but the name seems to sugest it was apropriate).
Seems to me it may be just a problem at high load (>29MB/s?) as the 3 way
raid that i did get to work was limited in speed as it used the slowest
drive, and the slowest drive was on a stable controller (via)
I figure its a known problem, but seems i did the tests i may as well share
the info.
I did have this problem a while back but destroyed my hpt366 controller
card, i just started playing with its replacement again.
Im not on the linux-kernel mailing list so could you please cc me if you
wish to discuss it further with me. (maybe theres another test i could do)
Thanks
Glenn McGrath
-
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/