|Subject:||[Gnumed-devel] pipelining experiment|
|Date:||Sun, 02 Oct 2005 23:53:23 +0800|
Some holiday fun while Karsten's away:-
the pyPgSQL module uses cursors so it takes I think 3 -4 round trips (create cursor. execute cursor, fetch result, close cursor) for every sql statement , so this slows down the access remotely salaam.
I used the script from test-area/sjtan/schemascan.py to retrieve a Kirk's record ( which is Id= 12),
and this took some time too. schemascan2.py reorganizes the single threaded single connection retrieval into
a multi threaded worker pool each with it's own connection, and uses a Queue.Queue to submit statements to available workers.
I was able to use about 30 threads; It looks faster, but using unix "time python schemascan2.py" the user time is still about a minute.
( Each statement will still take 3 round trips because of the use of sql cursor, although now there is some pipelining, due to concurrently running connections ).
How long does it take running as a batch file to psql ?
If the XMLRPCServer serving yaml, text, xml , serialized python object or whatever, did materialize, what would be the gain ?
( using a web server, one could retrieve a constructed record in 10 seconds , I think).
|[Prev in Thread]||Current Thread||[Next in Thread]|