Charles M. Hannum writes:

>> The fact that the directory on the RAID5 contains only subdirs
>> as entries, and in /dev it's device nodes shouldn't be of any
>> significance, imho?
>
>Actually, that makes a HUGE difference.  Directory inodes get spread
>all over the disk, so this will always be slower.

Hmm, yes.. haven't thought of that...

>Could you `dumpfs' your file system, so we can see the cylinder group
>layout?

The dumpfs output is rather large (3.5M uncompressed),
here's the top of it:


endian  little-endian
magic   11954   time    Tue May  1 01:59:55 2001
cylgrp  dynamic inodes  4.4BSD  fslevel 3       softdep disabled
nbfree  2045513 ndir    2010    nifree  4474995 nffree  13008
ncg     4375    ncyl    139970  size    17916160        blocks  17286075
bsize   8192    shift   13      mask    0xffffe000
fsize   1024    shift   10      mask    0xfffffc00
frag    8       shift   3       fsbtodb 1
cpg     32      bpg     512     fpg     4096    ipg     1024
minfree 5%      optim   time    maxcontig 8     maxbpg  2048
rotdelay 0ms    headswitch 0us  trackseek 0us   rps     60
ntrak   1       nsect   256     npsect  256     spc     256
symlinklen 60   trackskew 0     interleave 1    contigsumsize 8
maxfilesize 0x0000400801017fff
nindir  2048    inopb   64      nspf    2
sblkno  16      cblkno  24      iblkno  32      dblkno  160
sbsize  2048    cgsize  2048    offset  128     mask    0xffffffff
csaddr  160     cssize  70656   shift   9       mask    0xfffffe00
cgrotor 705     fmod    0       ronly   0       clean   0x02
insufficient space to maintain rotational tables


The full dumpfs output can be downloaded (gzipped, 100K) from
http://www.mukappabeta.de/pub/www/mkb/stuff/rraid1e.dumpfs.gz
I'm a bit troubled about the last line in the above text ("insufficient
space to maintain rotational tables") although I'm not sure what
it means exactly.

--mkb