Building A Better Map (cont.)
This morning I finished my "better map" for the instrument messages for my ticker plant. I had a few typos because I did a bit of copy/paste to get the three levels coded up as quickly as possible, and there was a lot of coding due to the extensive use of pointers to save space.
In any case. I got it to compile and then ran through a few simple tests. Fixed a few typos, got the tests to work, and then decided to try a speed test. The code was going to make 1000 messages, store them in the map, and time the results. The nature of the map was such that the put() method returned the old Message pointer for that instrument so that you could dispose of it, or make note of it, etc.
So the test loop looked like this:
msg::Message *oldie = NULL; msg::message::Print *print = NULL; uint64_t startTime = msg::TransferStats::usecSinceEpoch(); for (int i = 0; i < 1000; ++i) { print = new Print("S:IBM", startTime, (100 + i), 1.50, (10 + i), 'T', 0); oldie = m.put(print); if (oldie != NULL) { delete oldie; oldie = NULL; } } uint64_t elapsedTime = msg::TransferStats::usecSinceEpoch() - startTime; log.info("updated the S:IBM Print 1000 times in %ld usec", elapsedTime);
The response was more than I could have hoped for:
28-Oct-10 08:26:30.08 INFO imap : updated S:IBM Print 1000 times in 816 usec
Wow! That's better than once a microsecond! That's fast. Very fast.
Now I get to work it into the code and see how it impacts the overall performance. I know it'll be better than the conflation keys I've been using, and that's a great boost for the ticker plant.
Nice!