db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael.Gir...@objectweb.org
Subject Re: java1.4.2 "rws" mode fix: LogToFile.java diff
Date Fri, 03 Sep 2004 14:29:28 GMT
I have been developing a log facility specifically for Transaction Manager 

XA Commit protocol.  The project (howl.objectweb.org) has been working
toward a throughput target of 10,000 tx/sec to the journal.  I have tried
both "rw" with appropriate force() calls and "rwd".  What I found is that
the "rwd" mode is a little slower than "rw".  Difference depends on the
type of disk.  On my 7200 rpm ATA drive, the "rwd" is almost 40% slower
than "rw".  On my 10,000 rpm SATA drive, the "rwd" is only about 12%
slower than "rw".

The IO pattern is quite different for XA protocol than it will be for
database logging, specifically, there are only two log records per
XA commit, and the first one must be forced.

I will be interested is hearing how "rwd" does with the Derby log.

The HOWL logger will be used by Apache Geroninmo Transaction
Manager and is being used by ActiveMq.  There may be an opportunity
to apply lessons learned from our projects (Derby and HOWL) to
both log facilities.

Michael Giroux, Objectweb.org

Jan Hlavaty <hlavac@code.cz> wrote on 09/03/2004 05:57:05 AM:

> Steen Jansdal wrote:
> > You are probably right. I don't know what mode I'm trying since I'm 
> > using the derby code unmodified. I just trying to tell you what I'm
> > seeing with my file monitor. I'm not a derby expert, yet, so I don't
> > now all the small technical details.
> Yes, Derby code in org.apache.derby.impl.store.raw.log.LogToFile is 
> using "rws" mode heavily, and no "rwd" at all. I suggest we try "rwd" 
> and see if performance is improved.
> Jan

View raw message