[ENBD] 2.5.46 update

[Peter T. Breuer]

"A month of sundays ago Tad Kollar wrote:"
> > Can you observe the growth with time or requests of this figure?
> 
> After rebooting and restarting the clients and lvm, the size of the module is 
> 116k and holding steady, even after a large copy. Maybe it was the lvm 
> pvcreate process or format that aggravated it before, or else time is a 
> factor... it was up for about 24 hours. There weren't any noticable effects of 
> the huge module 'size'.

You'll have to try and catch it growing, to see what the trigger is.

Logically, it must be corruption. That number is meaningless.

> I tried merge_requests=30 and got a hard lockup during sync after a 213M copy 
> operation. Not 100% sure merge_requests was the cause, but without that option 
> the copy and sync succeeded on a second try. Performance seems okay... I guess 

merge_requests makes no changes in the driver itself. It merely changes
the "limit" that the driver tells the kernel that it prepared to accept
per request. The kernel is responsible for merging requests up till
that limit size (but no more) before the driver ever sees them, in the
elevator code.

So if you can get a lockup with merge_requests, and the problem is in
the driver, then you can get the lockup without merge_requests.

> I'll try a bonnie run.

OK.

Peter