On Sun, Apr 14, 2013 at 8:31 AM, someone <newsbo...@gmail.com> wrote: > Ok, thank you. I just came across a blog that said pytables is also a very > good option? > > http://www.pytables.org/moin/PyTables?action=AttachFile&do=view&target=non-indexed.png
>From what I gather, that's looking at performance of a non-indexable query on a 10,000,000-row table. That's going to suck whatever you do, and the exact level of suckitude doesn't really prove much. (Note that even the best options are taking half a second for this single query.) A better test of a database is transactions per second of something that approximates to your real workload. For instance, English Wikipedia has roughly a hundred edits per minute (assessed by me just now by looking at the Recent Changes), and some ridiculous number of page reads per minute (not assessed, but believed to be somewhere between 11 and Graham's number); so a test of a proposed new database would have to mimic this ratio. Most of the queries involved should be able to be answered using indexes; in some cases, ONLY using the index (eg if you just want to know whether or not a row exists). PyTables may well outperform PostgreSQL in real usage, but that one graph doesn't tell me that. (Not to mention that it's measuring a somewhat old PG.) ChrisA -- http://mail.python.org/mailman/listinfo/python-list
