| Hi Gj,
Yesterday I performed a test to trade fetch 100 available documents using
1 and 2 AFS processes.
Watching both processes via SHOW PROC/ID=../CONT showed that both processes were
active during the trade fetch, ie consuming I/O and CPU.
I didn't test if this also has an impact of the throughput. I'll try this today.
The maximum number of AFS processes is determined by the value as specified in
the field "Maximum number of extra Application processes" in the
V1.3 Support Services Parameter Details, via edit config ->Maintain server
parameters.
But I have to admit this is a strange place to specify the maximum and if you
fill in something in the "V1.3 Support Services Parameter Details" you have to
specify a value for "Copy failure timeout".
Martin
|
| Hi GJ,
It did some testing with 100 availble documents. Performing a Trade Fetch
with 1,2 and 3 AFS processes didn't hardly make any difference in the total
execution time. It even looks like more AFS processes increase the total
execution time.
100 documents fetched :
Number of AFS execution time
1 84 secs
2 85 secs
3 98 secs
tests have been performed during the day. The last test can be influenced by the
total system load.
After my opinion the overall execution time isn't dependent on the number of
AFS processes. If I'm wrong I'm sure someone will correct me.
Althought the help text doesn't give a clue how to stop additional AFS processes
, just type shutdown file....
Martin
|
| Running multiple AFS's could slow you down. The AFS holds a list of
AVAILABLE documents in its cache. This is much faster than always
checking the database, *but* the cache is private to the AFS, it is not
shared. If you have 2 or more AFS's they will have duplicate
information. In certain cases, an AFS could select a document which
another AFS has already delivered. This will not be visible to the
user, but the document will be extracted from the database only to find
that it is not appropriate and the AFS cache will be searched again.
Using a single AFS does appear to create a bottleneck, but in reality
it speeds up the fetching process. The AFS delivers only a pointer to a
document, it does not fetch it, so the slowest part (extracting the
document) is not part of the AFS.
We recommend using only a single AFS. Only the translators and
converters can be usefully replicated.
|