> On Mon, 21 Jul 2014 13:05:13 +0000, Markus Schaber <m.scha...@codesys.com> > said:
> Hi, > Von: David Canterbrie >> I've been tasked with trying to understand how much of a performance hit one >> would get if one had to scan a table in its entirety versus reading the same >> data stored as a new-line (or some sort like that) from a file. >> >> The hypothesis I suppose we're trying to understand is that reading >> sequentially from SQLite (without indices) should be comparable to reading >> from a file that has the same data +/- 1-2% >> >> My first question is that does sound reasonable, and has someone ever done >> such a test? > I guess this highly depends on your data format and parser code. > The author of > http://sebastianraschka.com/Articles/sqlite3_database.html#results > claims a factor ~20 speed advantage for SQLite. I read that page when if was first posted here a few days ago and found that claim extremely hard to believe (with those particular programs and that data), not the least because query_sqlite_db.py executes a query that never fetches anything from the database (the rows were inserted with feature1=Yes and the select query binds the parameter to "YES"). Just for fun... read_lines.py takes just over 1 second on a somewhat slower Xeon, for a 121MiB file containing 6M lines of the format ``this is line %07d\n''. As does query_sqlite.py when it's looking for "YES" and returning nothing. With the correct binding it takes about 10s. Warm cache in all cases. And a cpython program that simply increments a counter 6M times takes about 0.7s on the same machine. Regards, -- Stelios. _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users