On Sun, Jun 24, 2007 at 11:31:50AM -0700, Gerald wrote:
> Ultimately i'm trying to run an import on 23 LTO tapes which is going
> too slow and i'm trying to determine why. When it first started for
> the first 16 tapes or so, I was seeing 30MB/sec+ import speed. On
> Friday I noticed a lot of I/O activity on hdisk0 which is where the
> TSM DB is located and thought maybe i could achieve higher import
> speeds if I striped that JFS2 file system across both the internal
> hdisks. Thats where my troubles began..
>
> I managed to arrange this through a series of AIX commands creating a
> new striped fs and creating/deleting dbv's but when done, performance
> was worse. I put it back to how it was and performance has not
> recovered. I now tried putting the DB on a raw partition and still
> performance isn't any better.
>
> I'm not sure where to go from here. hdisk0 where the raw partition is
> located is still pegged (see below). A q stgpool takes quite a while
> to come back and ultimately i'm importing at a rate of about 6MB/sec
> now.
>
> The bad news is i'm on tape 17 of 23. Tapes were taking about 7 hours
> to import at 30MB/sec and i suspect we're now importing at a rate of
> 1/5 of that so lets call it 28 hours per tape. If true, the remaining
> tapes will take 7+ more days to import.
>
> My one idea is i could dump and load the DB to reorg it (perhaps thats
> causing the slowness?) but that would involve halting the import and
> starting it over - something that would also cost me 3-4 days or so.
> So it may be worth it but I don't even know if that's the root of the
> problem.
>
> Anyone have any ideas/suggestions?

what type of machine?  How many CPU's, how much memory.

OSlevel?  ML?

What about disk controllers, what type of disk is hdisk0?

Is your Log and DB on the same disk?

One thing I would do, is install 'nmon'.  Check to see if you are thrashing.

Are you seeing any disk errors?

Just some thoughts.

hth,

bob