Ok, Thanks Leo.
I agree with you but the operations are additive (and not in parallel), so
I'm trying to squeeze the maximum of log information for the same price :).
Logging will stay enabled during production but we want to be able to turn
it off/on/change level dynamically, so the if (LOGGING) would not work. The
"new" will instanciate about 10K logger objects per second and I was just
wondering if it would make any difference (especially by triggering the gc
more frequently maybe).
But I agree this is probably negligible compared to other stuff.
Thanks
-Vincent
----- Original Message -----
From: "Leo Sutic" <[EMAIL PROTECTED]>
To: "Avalon Developers List" <avalon-dev@jakarta.apache.org>; "Vincent
Massol" <[EMAIL PROTECTED]>
Sent: Friday, November 30, 2001 1:12 PM
Subject: RE: Loggers, final classes and performance
> Vincent,
>
> I think the overhead of one object creation is negligible in this case. If
> you log something, then you have:
>
> 1) Object creation.
> 2) Function call to logger.
> 3) IO operation (write the log entry).
>
> Of these three, number 3 is the overwhelmingly slowest operation.
>
> If you have your log statements inside your innermost loops, the consider
> wrapping them in a:
>
> if (LOGGING) {
> ...
> }
>
> where LOGGING is a static final boolean variable. I read somewhere that
this
> will have the same effect as conditional compilation.
>
> If you do not have log statements inside your innermost loops, the loop
> itself probably overwhelms the performance penalty of a maybe-log. Then
> again, if you do log something, and it involves disk access, it will
> overwhelm most non-I/O bound operations.
>
> In short - if you get performance problems, your problem is not the extra
> wrapper object, it is the fact that you are logging too much.
>
> /LS
>
> > -----Original Message-----
> > From: Vincent Massol [mailto:[EMAIL PROTECTED]
> > Sent: den 30 november 2001 10:16
> > To: avalon-dev@jakarta.apache.org
> > Subject: Loggers, final classes and performance
> >
> >
> > Hi,
> >
> > I need your expert advice here. The Log4JLogger class is declared
> > as a final
> > class. In my code, I use :
> >
> > Logger logger = new Log4JLogger(
> > Category.getInstance(targetName));
> >
> > This is used in a generic method that is called to automatically
> > log entries
> > and exits of methods, so it is executed every time there is a log
> > to be made
> > for an entry/exit. It is thus called very often.
> >
> > I'd like to know if it has any performance issue. More specifically, I'd
> > like to know if the final keyword helps for performance (as it does for
> > final methods which are inlined) : I guess that even with the
> > final keyword,
> > there is always a new object instanciated in memory ...
> >
> > So, I also guess that simply writing :
> >
> > Category category = Category.getInstance(targetName);
> >
> > would be more performant ?
> >
> > Another solution would be to have a hashmap of all loggers and
> > check in the
> > hashmap whether the logger for a given targetName already exist
> > and then use
> > that logger if it does. However, the additional work (putting the
> > instances
> > in the hash, possibly with synchronisation (argh!), then doing an "if"
and
> > then fetching the logger from the hash) might overall be more costly
that
> > instanciating a new object.
> >
> > What do you think ? Any idea ?
> > Thanks
> > -Vincent
> >
> >
> >
> > --
> > To unsubscribe, e-mail:
> <mailto:[EMAIL PROTECTED]>
> For additional commands, e-mail:
<mailto:[EMAIL PROTECTED]>
>
>
>
--
To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
