db-derby-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From <de...@segel.com>
Subject RE: Speed of using Derby DB
Date Thu, 19 Jun 2008 10:25:09 GMT
 

I haven’t looked at Derby’s code, but just because you haven’t ‘committed’
the transaction doesn’t mean that Derby isn’t writing the records.

Without a commit, depending on your isolation level, other users won’t see
or feel the effects of your data. The transactions are stored in a log so
that you can rollback the transaction, the data is still getting written.

 

(A simple test would be to write a record of a known length, X number of
times, on a machine where the JVM has a set restriction size. If you think
everything is in memory and not being written to disk, then after Y number
of iterations, you will fill up the heap and go boom.)

 

But Derby can be different so what do I know?

 

  _____  

From: vodarus vodarus [mailto:vodarus@gmail.com] 
Sent: Thursday, June 19, 2008 3:43 AM
To: Derby Discussion
Subject: Re: Speed of using Derby DB

 

Hi,

By using Øysteins approach I was able to get the time down to 2.4 seconds on
my machine, on which the client [1] and stored procedure code took around 12
seconds. The best I could get on the latter, tweaking page cache size and
page size, was around 8 seconds.

By cheating and removing some durability guarantees, I got down to a best
time (not quite stable) of 1.5 seconds using Øysteins suggestion.

I was surprised of the high disk activity seen when running the code. Lots
of writes are taking place, which I did not quite expect for Øysteins query.
But I do not know the implementation or the algorithm being used.

There also seem to be some overhead invoking a stored procedure, as the
client [1] code is faster. This would of course look different if the
network JDBC driver was used, as you wouldn't have to transfer the data over
the wire.

To me it seems what takes most of the time is updating the result table.

So in short, no fresh ideas! Anyone else?
I didn't try using batches for the updated though.


PS: Note that your pageSize setting is invalid (must be one of 4096, 8192,
16384, or 32768) and Derby will silently ignore it and use the default...


-- 
Kristian


[1] Note that client in this case still refers to the embedded driver, but
the code composing the stored procedure is invoked from the driver side
instead of "inside" the database.

Hello )))

I set pageSize to 32768, but result time seems near 11-12 sec. 


What is the "Øysteins approach "? Can you write steps to get 2.4 seconds
time?

Thanks.


PS "To me it seems what takes most of the time is updating the result
table." But what is the problem there? I commit data at the end, so DBMS
should not do any writes ...


Mime
View raw message