Re: Unicode database + JDBC driver performance

Jan,

You say you are using 7.2.1, is that for both server and jdbc driver?
There is a performance patch in the 7.3 driver that bypasses the built
in java routines for converting to/from utf8 with a custom one.  The
built in java routines are very slow on some jdks (although on jdk1.4
they are pretty good).  Can you try the 7.3 drivers?

thanks,
--Barry


Jan Ploski wrote:
> Hello,
>
> I have some questions regarding PostgreSQL handling of Unicode databases
> and their performance. I am using version 7.2.1 and running two benchmarks
> against a database set up with LATIN1 encoding and the same database
> with UNICODE. The database consists of a single "test" table:
>
> Column |  Type   | Modifiers
> --------+---------+-----------
> id     | integer | not null
> txt    | text    | not null
> Primary key: test_pkey
>
> The client is written in Java, it relies on the official JDBC driver,
> and is being run on the same machine as the database.
>
> Benchmark 1:
>
> Insert 10,000 rows (in 10 transactions, 1000 rows per transaction)
> into table "test". Each row contains 674 characters, most of which
> are ASCII.
>
> Benchmark 2:
>
> select * from test, repeated 10 times in a loop
>
>
> I am measuring the disk space taken by the database in each case
> (LATIN1 vs UNICODE) and the time it takes to run the benchmarks.
> I don't understand the results:
>
> Disk space change (after inserts and vacuumdb -f):
> LATIN1      UNICODE
> 764K        640K
>
> I would rather assume that the Unicode database takes more space,
> even 2 times as more.. Apparently not (and that's nice).
>
> Avg. Benchmark execution times (obtained with the 'time' command, repeatedly):
> Benchmark 1:
> LATIN1      UNICODE
> 11.5s       14.5s
>
> Benchmark 2:
> LATIN1      UNICODE
> 4.7s        8.6s
>
> The Unicode database is slower both on INSERTs and especially on
> SELECTs. I am wondering why. Since Java uses Unicode internally,
> shouldn't it actually be more efficient to store/retrieve character
> data in that format, with no recoding? Maybe it is an issue with the
> JDBC driver? Or is handling Unicode inherently much slower on the
> backend side?
>
> Take care -
> JPL
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 3: if posting/reading through Usenet, please send an appropriate
> subscribe-nomail command to majordomo@postgresql.org so that your
> message can get through to the mailing list cleanly
>