On Wed, 8 Feb 2017, Dimitris Bil wrote:
Do you perform the benchmark on the native database table using cold
cache or warm cache? Also, can you briefly describe what the
benchmark does and give the schema for the native database table?
My benchmark repeatedly reads all of the columns one by one given row
id and column name. The table is read many (e.g. 100) times so this
is a warm cache test.
The schema is not terribly important but the table we are trying to
optimize (with 1800 or less rows) contains a 64-bit rowid, five
integer values, and two short text string values.
int64, uint32, uint32, text[16], uint8, text[16], text[18], uint8,
uint32
What I am looking for is expected average virtual table performance vs
native table performance for repeated column reads.
Due to being a generic implementation (supporting many virtual
tables), our virtual implementation uses programmed/dynamic
marshalling rather that compiled marshalling. The schema definition
is also dynamically generated.
There are implementation overheads and it is useful to know what
performance is possible (e.g. compared to native table performance)
in order to know when the implementation is about as good as it can
be.
Bob
--
Bob Friesenhahn
[email protected], http://www.simplesystems.org/users/bfriesen/
GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
_______________________________________________
sqlite-users mailing list
[email protected]
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
