Re: test request

[Thierry Deval <TDeval@PrimeOBJ.COM>]

On 31-Mar-00 Thierry Deval wrote:
> ...
> 
> Backup -> Snapshot install -> Reconfigure Everything

Voila, the clean snapshot install is done.

It went as a breeze. (this won't surprise anyone ;-)
This was done happily with the new adaptec driver.
(Yes, Steve, all was fine for this one disk install :-)

Even, now that my network is up and running again, I can use it as my
ADSL NATted router, as before.

But the "Reconfigure Everything" part will not be done soon... :-(

Indeed, as soon as I configure my RaidFrame set (where my home, src,
and data partitions reside), the system soon becomes unusable.

I explain :

  With the patch Steve provided me, I can configure the RAID, fsck and
  even work on it a little. (the writes are fine now)

  The problems arise when I do something involving a big load on the
  disk read; for example a tar tvf <some_big_tar>.
  After some time (1-2 sec), the command locks completely. It won't
  even respond to a kill -KILL.
  Even if the read doesn't happen on a RAID disk. (yes, it happens with 
  the just installed sd0 disk)

  The strange part is, while this locking happens, other commands still
  work, and the network is fine though with degraded performance. (a
  ping which usually gives me ~200ms jumps to ~5000ms !!??  Well, I
  still have to confirm this )

I would like to provide some more info about this problem, but when I
do a ktrace on the locking commands (I tried several), they complete
without any problem, with a huge ktrace.out . It would then hang if I do
the kdump on that more than 200M ktrace file !!!

This sounds like a timing problem. But why does it only happen when
RaidFrame is used ? That same RaidFrame enabled kernel, when used as
plain GENERIC, doesn't present the problem...

So, what should I do ?
Use a kernel before the Adaptec upgrade (and snapshot), or convert back
my RAID system to a normal disk distributed config ?


T.