remo strotkamp wrote:
> Hi raider out there,
>
>
> We kept getting error messages under 'heavy' load ( RAID resync, e2fsck,
> copying
> a 2GB directory tree..) which looked as follows:
>
> Jan 21 16:42:45 freshkills kernel: hdk: dma_intr: status=0x51 {
> DriveReady SeekComplete Error }
> Jan 21 16:42:45 freshkills kernel: hdk: dma_intr: error=0x84 {
> DriveStatusError BadCRC }
>
> Most of the times things just went fine, but sometimes the corresponding
> harddisk
> received a DMA reset:
>
> Jan 19 21:16:39 freshkills kernel: hdk: DMA disabled
> Jan 19 21:16:39 freshkills kernel: ide5: reset: success
>
>
I have been having these sorts of problems for a fair while, my opinon (and
thats all it is) is that raid cant handle much more than 40MB/s,
irespective of your setup, i think theres some other limiting factor
Ive gotten ~35MB/s with 3 udma33 drives, each its own channel, i got no
timeouts.
When i reconfigured it to 4 drives i saw a small performance increase, but
the results were suspect as i could test with large file sizes as i would
get timeouts etc (as your seeing).
Actually now i think about it, just yesterday i got timeout errors under
heavy load just copying between non-raid partitions (copying 3 GB from one
drive to another) , the CPU usage is pretty high as well.
Do you ever get timeouts other than in raid partitions ?
Id be happy is it was my cables (as you suspect your problem is) but im
sceptical.
I guess youve read the warning in the kernel archive about how using hdparm
can cause these sorts of problems.
Im using the HPT366 controller b.t.w.
Im thinking high cpu usage could be the problem, mine is a k6-2 266, had
128 MB ram when doing raid tests, but only 64 yesterday when i got timeouts
xfering between partitions.
There should be hardly any cpu usage for the controllers as they use DMA..
i guess your case of raid5 has high cpu overheads.... I cant explain my
situation
Wish i had a solution for you, let me know if you find one.
Thanks
Glenn McGrath
P.S. Im sure you realise that if ou got timeout errors when running bonnie
then your results dont mean anything.
>
> Now via deja.com, we narrowed the cause of the problem down to
> cabling!!!!!!
> Though we use the special cables they have to be too long, as the
> standard rack-mountable
> case is too big with PCI-slots in the back and HD's in the front.
>
> And the fact that the HD-cases use som additional cable inside and add
> another connector
> is not very helpful neither!!!!
>
> So for the moment we use normal cables again, which makes the disk fall
> back to UDMA33 and
> the errors seem to be gone!!!!!!! :-)
>
> But in the end we would prefer the UDMA66 back, especially because of
> the block-read performance loss
> of 1 third....
>
> So what we need is a smaller rackmountable case.
>
> And we wondered whether there are some raiders out there which
> successfully use
> UDMA66 RAID in a rackmountable case and with HD-cases and what
> rackmountable
> case they are using......
>
> I thought maybe somebody would be interested in a little 'linup' of
> UDMA66 versus UDMA33...:-)
>
> Both are for a RAID-5 system with 5 disks, done in single user mode.
> RAM is 128MB and its a single PIII 500Mhz if I am not mistaken....
>
> Interestingly writes do not seem to suffer but block-writes pretty
> heavily...
> And that with 1HD/ UDMA66 bus!!!!!!
>
> Wasn't there once a mail from somebody saying that todays HD's wont fill
> even
> the UDMA33 capacity....
>
> -------Sequential Output-------- ---Sequential
> Input-- --Random--
> -Per Char- --Block--- -Rewrite-- -Per Char-
> --Block--- --Seeks---
> MB K/s %CPU K/s %CPU K/s %Cpu K/s %CPU K/s
> %CPU /s %CPU
> :::::::::::::first run UDMA66
> 2047 7184 90.2 28856 23.2 17191 32.5 8301 97.5
> 60630 43.0 201.1 3.7
> ::::::::::::::second run UDMA66
> 2047 7183 90.2 29521 24.1 17086 31.7 8305 97.5
> 60009 43.3 201.6 2.8
> ::::::::::::::UDMA33
> 2047 7153 90.0 29793 24.1 14619 26.6 8250 96.9
> 40113 27.9 161.3 2.8
>
> anyway,
>
> would be nice if people could respond with their rackmountable cases
> information...
>
> and:
>
> did anybody get the above mentioned errors and crashes even with short
> enough
> cables and no HD-cases?????
>
> have fun raiding
>
> remo
