I wasn't clear on whether the client in question was AIX, but if so,
one of my coworkers has been working with IBM support for a couple of
months on a performance problem related to a large number of files.
IBM has recently stated that they do not intend to fix the problem.

AIX maintains an inode cache in memory.   The size of the inode cache
is based on the amount of memory in the system, and is not tunable.
The problem occurs when the inode cache fills (because of accessing
enough files to fill it).   Once you fill the inode cache, inode access
becomes very slow and the CPU is pegged.

I don't know the exact formula, but my coworker did some tests with
increasing numbers of empty files.   He accessed them by doing:
"find /whatever -print".   The find was fast as long as the number
of files was less than the number of inode cache entries.

When we reached the magic number of files (seems like it was about
30000 on a 256M system, but don't quote me), run time and CPU went
through the roof.   Fortunately, we don't back up the system in
question with ADSM :)

--
Gene Mangum
Gene Mangum
University of Michigan Medical Center


On Wed, 6 May 1998, Thomas A. La Porte wrote:

> If there is a maximum (which I do not believe that there is),
> 90,000 is probably not it. We have several filesystems in the
> 100,000 - 200,000 file range. You might look at an
> INSTR_CLIENT_DETAIL trace on an incremental backup of the
> filesystem. It will help you assess where ADSM is spending all of
> its time.
>
> Using this type of trace, we were able to determine that for one
> file server we had a fixed overhead of nearly 3.5 hours for an
> incremental backup. If we only backed up 1MB from that filespace,
> it would still take 3.5 hours, due to the overhead involved in
> parsing a directory structure of so many files, and determining
> whether or not ADSM needed to back them up.
>
>  -- Tom
>
> "It's a dog eat dog world,                Thomas A. La Porte
>  and I'm wearing milkbone underwear."     DreamWorks SKG
>               - Norm Peterson             <tlaporte AT anim.dreamworks DOT 
> com>
>
> On Wed, 6 May 1998, Cohn, Grant wrote:
>
> >Greetings
> >
> >Is there a maximum number of files that can be backed up by ADSM in one
> >file-system?
> >
> >We have a file-system with approx. 90 000 data files in it, and while
> >doing an incremental backup of this file system, ADSM always get to the
> >line : ANS4102I ***** Processed    18,500 files ***** and then stops.
> >
> >After the tcp-timeout period (I have set this to two hours!) we get the
> >message :
> >
> >               ANS4017E Session rejected: TCP/IP connection failure.
> >
> >The same thing happens every night during our scheduled backup AND also
> >happens if I do an INCR backup from the dsmc> prompt.
> >
> >The scheduler then says it will try again in 20 minutes, after which
> >time the window period has elapsed.
> >This results in this particular file-system failing and all subsequent
> >file-systems not to be backed up.
> >
> >We are running ADSM Server 2.1.5.15
> >                                  Client 2.1.10.7
> >On AIX RS/6000 J30. AIX 4.1.5
> >The Tape drive is an IBM 3570.
> >
> >This problem is happening on the server that has the 3570 attached to
> >it! (Is ADSM still talking via TCP/IP in this case?)
> >
> >Any suggestions / ideas to try would be most welcome!!
> >
> >Many thanks
> >
> >Grant Cohn
> >Shell & BP Petroleum Refineries.
> >Durban, South Africa.
> >e-mail : grant.cohn AT sapref DOT com
> >
>