Re: Time to heat up the new module discussion

[Ben Maurer <bmaurer andrew cmu edu>]

On Fri, 14 Jul 2006, Nate Nielsen wrote:

> Steve Frécinaux wrote:
> > Iain * wrote:
> >
> > > > As for .NET, even Microsoft themselves had to pull back from using it for core
> > > > functionality due to performance reasons - why do we think we will do any better?
> > > As someone who is running mono based applications fairly regularly, I
> > > haven't noticed any major performance issues. We're not talking here
> > > about replacing the core libraries with c# based ones, we're talking
> > > about applications, and for me the mono based apps are just as fast as
> > > the C based ones.
> >
> > I'm not really against having C# apps in the core (in fact I don't
> > really mind), what I'm more frightening about is having applications
> > that run all the time, using managed languages, and, as a consequence,
> > taking up a fairly large amount of memory, from the computer start to
> > the shutdown.
> >
> > That's what I'm against including libraries, daemons or applets written
> > either in C#, or in python, while having small apps you close once
> > you're done (like alacarte) is fine.
>
> Perhaps daemons, applets and core libraries should evaluated on their
> performance merit independent of the language they're written in or the
> framework that they use.
>
> I'd imagine that those who have worked so hard on GNOME performance
> issues would be disappointed if they saw a C applet take 18 - 20 megs
> (resident) or take 10 seconds to load.

PLEASE learn how to measure memory. Resident is *NOT* a good measure of 
memory. You must use smaps to even begin getting an accurate measurement. 
Not a single person other than Federico and myself have used a number 
about performance that is relevant. This suggests that there are many 
people `me too'ing and not thinking about what they are saying. If you 
want to make groundless claims, comment on slashdot.

> If the GNOME desktop becomes dependent on one single process like that,
> it wipes out most of what so many people have worked hard step by step
> to gain as far as startup speed and memory usage.

As you might realize, I'm one of those people. IMHO, having applications 
written in a framework that encourages maintainable, clean code is a step 
forward for performance. Do some of the Mono based applications need 
profiling: yes. Should we use performance as an excuse for not having 
Mono: no.

> I don't think we should cut slack for a particular applet or daemon
> because it happens to be written using an inefficient platform. If its

I did not hear anybody suggest that. We're not cutting slack for anyone.

> performance issues are not up to snuff, then it shouldn't be included as
> part of the GNOME desktop.

Should it be included by *default*: probably not. It needs to be clear 
that Mono applications will be accepted without bias as to their language. 
This is what is being requested.

> That said, I bet the Mono guys could prove everyone wrong by optimizing
> managed code to where it approaches the startup speed and memory usage
> of lower level frameworks. That would be lovely. But it might mean not
> trying to emulate every last 'innovation' in MS .NET C# 3.0 etc... :)

Please show benchmarks that show issues here. Banshee and other c# apps 
seem to startup very fast. While memory usage isn't as good as it could 
possibly be, we're not talking about a huge amount here (see numbers from 
myself, federico). Again, in the long term, a moving GC will help us 
return memory to the system.

-- Ben