IDAD0 rebuilding (was: Re: camcontrol and IDAD0)

[tonix (Antonio Nati)]

At 18.58 09/09/2005, Matthew N. Dodd wrote:
>On Fri, 9 Sep 2005, tonix (Antonio Nati) wrote:
>>Finally, these controllers seems to be unusable in a production 
>>environment with FreeBSD.
>
>I don't think anyone has really tested rebuilding a volume while the 
>system is running.
>
>I'd be interested in the results of changing the definition of 
>IDA_QCB_MAX from 256 to 128.
>
>See sys/dev/ida/idavar.h:
>
>         #define IDA_QCB_MAX     256

With IDA_QCB_MAX => 128;

processes working
         process 1 -> tar of whole disk to file, then gzip file, then 
delete file
         process 2 -> rebuild world
         process 3 -> every ten second "idacontrol show"
action:
         during these jobs, I extract one disk, wait 20 seconds and 
insert it again observing until rebuilding has finished.


With DL360 G1 2x933 CPU and Integrated Smart Array, RAID = 1 (2 x 18GB disks)
No problems with controller rebuilding priority on every level (low, 
medium, high)
idacontrol always shows correct status and percent of rebuilding 
still to be done
Only some console logs like "ida_command: out of QCBs"
When disk is removed or inserted great bunchs of soft errors.
A lot of soft errors every time "idacontrol show" is run.
With DL580 G1 4x700 CPU and Integrated Smart Array, RAID = 0+1 (4 x 36GB disks)

         Systems never hangs but rebuilding does not work (apparently).

         When reinserting the extracted disk, rebuild does not start 
until I run "idacontrol recover".
         After running this command, rebuilding seem extremely slow, 
and after a few minutes
         status of RAID set turns to "interim". Again "idacontrol 
recover" and again "interim" after a few minutes.
         Display of rebuilding missing start from 50% (as only one 
disk of two must be rebuilded).

With DL580 G1 4x700 CPU and Integrated Smart Array, RAID = 5 (3+1 x 9GB disks)
         When Smart Array is initializating the RAID 5 set, bootstrap 
says Smart Arrays is yet optimizing datas,
         while idacontrol says arrays is OK. Who to trust?

         Still to test rebuilding with processes doing I/O.

Tonino

>--
>10 40 80 C0 00 FF FF FF FF C0 00 00 00 00 10 AA AA 03 00 00 00 08 00