Chuck,
I would suggest looking at an instrumentation client detail trace
(-traceflags=instr_client_detail ) for an incremental backup of
the filesystems. This will give you a good indication of where
the ADSM client is spending the bulk of its time. The following
is an example of the trace information you'll receive:
Final Detailed Instrumentation statistics
Elapsed time: 26.908 sec
Section Total Time(sec) Average Time(msec) Frequency used
------------------------------------------------------------------
Client Setup 0.679 678.9 1
Client Setup 0.679 678.9 1
Process Dirs 11.204 20.9 537
Solve Tree 0.000 0.0 0
Compute 0.013 0.0 520
Transaction 1.370 0.6 2206
BeginTxn Verb 0.000 0.0 1
File I/O 10.968 17.5 627
Compression 0.000 0.0 0
Data Verb 1.459 2.8 520
Confirm Verb 0.058 58.4 1
EndTxn Verb 1.129 1129.3 1
Client Cleanup 0.027 27.0 1
------------------------------------------------------------------
During troubleshooting of a V2 problem a while back, these traces
During troubleshooting of a V2 problem a while back, these traces
were key in showing that our disk subsystems were our primary
bottleneck.
-- Tom
Thomas A. La Porte
DreamWorks Feature Animation
tlaporte AT anim.dreamworks DOT com
On Mon, 10 May 1999, Chuck Mattern wrote:
>I have an AIX server (ADSM client) that has some large filesystems (~80 gigs)
>and a large number of files (~3 million). The machine has 512 megs of RAM, 1.5
>megs of swap and, until the number of files topped 1 million per filesystem, no
>performance problems. File average 40k in size. ADSM is used for backup and
>for HSM. Currently backup of one of these big filesystems can take as much as
>60 hours. We are forced to use memoryefficientbackup lest dsmc fail with an
>out
>of memory error. The machine never runs out of memory so it must be the dsmc
>client. We have tried setting the priority to the max with nice and have tried
>setting ulimit for memory and data to unlimited. Still must use
>memoryefficientbackup and still need hideous amount of time to backup.
>
>Has anyone dealt with a situation like this in the past (hopefully with
>success)?
>
>Thanks in advance,
>Chuck
>
|