[mnet-devel] Re: [mnet-cvs]this patch reverts calculate_timeout() to mu + 6*sigma; I ...

[Arno Waschk]

okay, but there is no SOMETHINGELSE*sigma which is right for everything 
either...
my patch was designed never to generate shorter timeouts (except for 
lowering the sigma factor...)
compared to the big sigma factor you put the effect of my assumed 500 for 
SOMETHING would be pretty small for small messages.
But i start getting tired from that theoretical discussions...
I *saw* those relay server logs....
Arno

On 18 Feb 2004 06:23:26 -0500, Zooko O'Whielacronx <zooko at zooko.com> wrote:

>> > i would still urgently vote for something like ((mu*size/SOMETHING)
>> > + 6*sigma). I definitely observed that this helped *a lot* in the 
>> relay
>> > server activities. And it does not really change the behaviour of 
>> small
>> > initiating messages. Arno
>
> By the way, I agree with you that it can help with "pass this along".
>
> I don't agree that it doesn't really change the behavior of small 
> messages.
> That's because of the SOMETHING.  If the SOMETHING is too big, it will 
> cause
> every message to timeout, and that will keep happening forever.  If the
> SOMETHING is too small, it will cause certain user-visible operations 
> like
> search and download to go slower because the timeouts are too long.
>
> Furthermore, there is no SOMETHING that is right for everyone.  You 
> might tweak
> it to be just right for your box, but then when you put it in the 
> software and
> ship that software to someone else, it is too big for his box, and 
> causes every
> message to timeout.
>
> A message-length-scaled dynamic timeout which calculated linear 
> regression would
> be free of this problem.
>
> Regards,
>
> Zooko
>
>
>
> -------------------------------------------------------
> SF.Net is sponsored by: Speed Start Your Linux Apps Now.
> Build and deploy apps & Web services for Linux with
> a free DVD software kit from IBM. Click Now!
> http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
> _______________________________________________
> mnet-devel mailing list
> mnet-devel at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/mnet-devel
>



-- 
http://www.arnowaschk.de


-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel