raid3 is slow

[Ivan Voras]

Szabó Péter wrote:

>>> After a few minutes

>  PID USERNAME  THR PRI NICE   SIZE    RES STATE    TIME   WCPU COMMAND
>    4 root        1  -8    0     0K     8K -       42:47 53.27% g_down
> 15400 root        1  -8    0     0K     8K -       82:18 19.38% g_bde 
> raid3/nmivole
>   11 root        1 171   52     0K     8K RUN    968:44 12.60% idle
>    3 root        1  -8    0     0K     8K -       22:12  3.03% g_up
> 19970 root        1 -16    0  3400K  1772K wdrain   1:12  2.93% dd
>   40 root        1  -8    0     0K     8K r3:w1   36:18  1.07% g_raid3 
> nmivol

> It seems terible!!! g_down gets more and more and more WCPU, and the 
> load goes to the skies. I don't know, what should i think. Now i can 
> see, at the end i will change the raid3 to stripe ;) But i don't want to :(

Also one of the "shouldn't happen" things - g_down is (roughly) the 
kernel thread that dispatches IO requests from the file system to the 
lower level devices, gbde and graid3 in your case. It's strange that it 
consumes that much CPU - it may mean that one of the GEOM classes is 
being unusually slow in offloading requests to its own thread, or that 
there's an unusual number of small requests bouncing between them. Try 
running 'gstat' during this test and post some results.

If you have the time, also try using geli instead of gbde, just to 
experiment.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 249 bytes
Desc: OpenPGP digital signature
Url : http://lists.freebsd.org/pipermail/freebsd-geom/attachments/20070329/9eccc389/signature.pgp