[Ocfs-users] Re: Clarification on 1.0.14-1

[Sunil Mushran]

Have you tried archiving to another ocfs volume?

The reason I bring this up is because you can avoid flooding
the buffer cache by using ocfs (o_direct) instead of ext2 (buffered).

That could be one possible reason for the slowdown.

Bruce Holzrichter wrote:

>>BTW, as this bug was known to cause on-disk corruptions, I am
>>not sure how you have linked this to your momentary freeze-ups.
>>    
>>
>
>Thanks, this sounds like it wouldn't be our issue then, as we are using
>OCFS for the DB and archive to an EXT2 partition.
>
>  
>
>>1. How many logfile groups do you have?
>>    
>>
>
>4 logfile groups on each node - 1 logfile per group
>Node 1 uses thread 1, node 2 uses thread 2 ...
>
>  
>
>>2. Logfile sizes?
>>    
>>
>
>Each logfile is 1 gig.
>Logfiles are on ocfs raid 0,1 and are archived to ext2 raid 5
>
>  
>
>>3. Kernel version?
>>    
>>
>
>Right now, Kernel 2.4.9-e.49enterprise
>
>  
>
>>4. Memory?
>>    
>>
>
>4 Node RAC, 4 Zeon CPU Dell 6650 16GB each.
>
>  
>
>>5. How did you determine the freezeup was during archiving?
>>    
>>
>
>This is the only common thread I have found so far, when one node
>archives, there is significant response slowdown from the DB. If more
>than one node is archiving at the same time, there is a tremendous
>slowdown, and our webservers back up traffic to the point our site
>sometimes fails.
>
>I am not the DBA, but the Systems Administrator.  We had been working
>with Oracle/Red Hat, and done numerous VM tuning changes, OCFS upgrades,
>kernel upgrades, etc along the way.  
>
>I have been looking at load issues, VM tuning, hardware, etc. trying to
>isolate where the issue lies.   These will even happen during light load
>on the systems when they archive as well, it appears.  
>
>Thanks for your thoughts on this.
>Bruce Holzrichter
>
>
>
>  
>