Re: bonnie++: disastrous RAID 1/5 results
martin f krafft wrote:
> also sprach Juri Haberland <juri@koschikode.com> [2003.08.21.1034 +0200]:
>> The first thing I notice is that you compare two different
>> filesystems: reiserfs and ext3. It is a known fact that reiserfs
>> is in most workloads *much* faster than ext3.
>
> Right, but while I had zero dataloss on ext3 filesystems so far,
> I have been screwed over by reiserfs a couple of times.
I didn't suggest that you actually *use* reiserfs, but if you're doing
benchmarks you should do them with the same filesystems, or else your
numbers are meaningless.
>> How are your three (four) disks connected? Do they all have
>> a seperate IDE channel or do some of them share one?
>
> The disk on the fast machine is master to a slave disk.
> The first disk in the RAID is (unfortunately) master to a slave
> CD-ROM, which is, however, unused.
Ok. I just saw, that I could have gathered that information from your
other mail - sorry about that.
>> Do all disks have DMA turned on?
>
> This might be the problem. I turn DMA on, always, in /etc/inittab.
> But it doesn't work. Check this:
>
> ailab:~# hdparm -d1 /dev/hda
>
> /dev/hda:
> setting using_dma to 1 (on)
> HDIO_SET_DMA failed: Operation not permitted
> using_dma = 0 (off)
>
> So what's going on here???
Your kernel doesn't support your IDE chipset. What kind of IDE chip (or
southbridge) dou you have on the slower machine?
You most certainly didn't compile or load the appropiate IDE driver.
>> Try with the defaults values for chunk size, algorithm and ext3 stride.
>
> Which are? I could not see any.
Well, I don't know about ext3, but my RAID5 device, which I created
without any special commandline parameters, looks like this:
md7 : active raid5 hdg9[2] hde1[1] hda9[0]
160086272 blocks level 5, 64k chunk, algorithm 2 [3/3] [UUU]
>> Use an external journal for ext3 that is located on a RAID1.
>
> Does this really increase performance if the RAID1 would be on the
> same disk(s)? I note that RAID1 give absolutely no speed improvement
> on writes. RAID5 does a little. And journals are more written than
> read, no?
It wouldn't really increase the speed if they were on the same disk,
though XFS (I know you use ext3) had a serious performance problem with
RAID5 and putting a external journal on a RAID1, even on the same disks,
solved that. This is fixed nowadays.
I'd say, fix your DMA problems and your RAID5 should fly.
Cheers,
Juri
Reply to: