I just ran some tests to determine the amount of time that would be required
to restore the files on my sites mail repository. I used
the -virtualnodename
option to enable a test system to restore files backed up from the
production
mail repository. Both of these systems were HP-UX 10.20 systems running
3.1.0.6 client code. The server was at 3.1.2.20 and ran under MVS. The
restores took about twice as long as the initial backups of the same
filesystems. I used the -traceflags=instr_client_detail options with two of
the restores. In each case, just over a gigabyte of compressed data was
transferred, and just over 200,000 files were restored. In each case, the
statistics generated in response to the trace flag attributed about three
quarters of the total elapsed time to 'Transaction', with a bit over a
million
occurances and an average time of about seven milliseconds per occurance.

I don't understand why there are so many transactions. We have the
txngroupmax
option on the server set to 256, and the txnbytelimit option on the client
set
to 25600 (indicating 25 megabytes). With these option values I would expect
something like a thousand transactions rather than a over a million. Can
anyone enlighten me on this point?

The average time per transaction is about the same as the typical round trip
time between client and server reported by the ping command. Does this
reflect
a real relationship between the two time measurements, or is it just a
coinicidence?