[Exim] Re: Dynamically Loadable Modules

Page principale
Supprimer ce message
Répondre à ce message
Auteur: Jason Robertson
Date:  
À: exim-users
Anciens-sujets: Re: Dynamically Loadable Modules (was Re: [Exim] Is Exin SNMP Aware? )
Sujet: [Exim] Re: Dynamically Loadable Modules
Actually what would be nice, though outta of scope for us, for awhile.. would be like the DLL
option, that Windows does do, would be to have preloaded libraries, so they are in memory only
once.

But as for another option, dynamically loaded modules could make exim faster, in that it would
also mean, a module would only have to load what it needs to run, for example if you are sending
email, why load the code to receive email, if it will never get used.
The other benefit is that it can reduce memory requirements of exim, as you can load and unload
modules as needed.

Jason

On 30 Aug 2001 at 1:53, Theo E. Schlossnagle wrote:

Date sent:          Thu, 30 Aug 2001 01:53:11 -0400
From:               "Theo E. Schlossnagle" <jesus@???>
Organization:       OmniTI, Inc. -- Computer Consulting
To:                 Richard Welty <rwelty@???>
Copies to:          exim-users@???
Subject:            Re: Dynamically Loadable Modules (was Re: [Exim] Is Exin SNMP Aware? 
)


> Richard Welty wrote:
> >>That's probably the biggest head-ache. I'm not sure of many systems
> >>where dynamic linking, on any level, could be considered "cheap" when
> >>done incredibly frequently, say on a dedicated mail server.
> >
> > i have to vote against the whole idea; it would be way too expensive
> > in a busy mail server.
>
> Where is your emprical evidence for this. I don't buy it. Dynamic linking
> happens all the time on my busy systems and things fly. Sounds like your
> dynamic linker is broken.
>
> Besides, most modules loaded into exim would be very small. So you are only
> looking at a mmap() and a symbol table verification. There are probably only a
> handful of symbols in one of these modules, so this overhead is TINY.
>
> I would rather load 100 modules than call {fsync,fsyncdata} just once on my busy
> systems. It the battle for speed, choose your bottleneck battles wisely.
>
> --
> Theo Schlossnagle
> 1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
> 2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7
>
>
>
>
>



---
Jason Robertson                
Network Analyst            
jason@???    
http://www.astroadvice.com