Thursday, September 25, 2008

Re: [JDBC] Performance comparison to psql.

I seem unable to perform basic calculus today... sorry again, with psql -1 the process took 97m
just to set the record straight -- the overhead from the implicit transactions was 15m

M.

On Thu, Sep 25, 2008 at 7:12 PM, Michael Nacos <m.nacos@gmail.com> wrote:
right! the -1 flag...

I repeated the psql test, this time with the -1 flag, and the process completed in 107m
so the transaction overhead in the previous psql tests is 5m
psql is probably reading each line from the input file and immediately submitting it
pgBee groups together many lines and batches them off to the server in one step

btw, it's operations/sec, not milliseconds in my previous email -- sorry! I am not using
prepared statements as I have to cope with arbitrary SQL, so it looks like I'm approaching
the performance of unbatched but prepared JDBC statements mentioned in this table:


Comparison table (records inserted per millisecond)
COPY JDBC JDBC batch
WITHOUT INDEXES: 198 1.5 14
WITH 2 INDEXES: 45 1.5 10

898 operations/second vs. 1500 records/second in the table above. Besides, these numbers
must be hardware-specific (I'm using a laptop with a 5400rpm disk)

Michael


No comments: