Re: Followup: opinions on Search services
- From: Miguel de Icaza <miguel ximian com>
- To: Jamie McCracken <jamiemcc blueyonder co uk>
- Cc: Joe Shaw <joeshaw novell com>, gnome-devel-list gnome org, Manuel Amador <rudd-o amautacorp com>, "John \(J5\) Palmieri" <johnp martianrock com>
- Subject: Re: Followup: opinions on Search services
- Date: Wed, 06 Apr 2005 22:01:00 -0400
> > This is a misnomer. C code can be every bit as inefficient and where
> > Python and C# are inefficient one can always go down and write modules
> > in a lower level language. Point is don't optimize until you need to.
> Actually mixing unmanaged with managed code is a real performance killer
> for mono. ANyway, the point was more to do with memory usage and in
> particular how garbage collected languages are currently quite poor for
> implementing this - see the beagle webpage about memory usage and its
> problems in this instance.
Do you have data to back out your assertion on "real performance
Transitions from unmanaged to managed code are not as fast as the
regular 5-7-instructions required on a call operation, but they are
under 20 instructions for any API that does not require extensive
marshalling (for example transforming an XmlDocument back and forth).
Not any worse than JNI for instance.
So your portrayal of "real performance killer" is slightly off-bases.
We -Mono people- consider that if you can avoid the transition to
unmanaged code you should do it by all means. But we ourselves use it
extensively everywhere in our base class libraries, and our compiler
(which uses pieces that transition a few million times back and forth
during a bootstrap) can compile 30,000 lines of code per second.
So your statement about "real performance killer" should probably be
qualified. As a general purpose statement, it is as precise as stating
that "good stuff is better than bad stuff". Well, duh.
] [Thread Prev