[mnet-devel] Re: [mnet-cvs]kb-scaled timeouts

[Zooko]

> > It isn't really buggy the way it is, but len(pickle(msgbody))
> > is not *exactly* the same as len(mencode(msgbody)).
> 
> its much closer to the real message size than assuming all
> messages to have the same size. ;)

I agree with *that*, but the size of the request message is not a good indicator 
of the turn-around time for the whole transaction, since it gives no indication 
of the size of the response message (which is often many orders of magnitude larger).

> no exceptions, just not working communications thanks to
> throttler and timeout estimations.

Do you have evidence that timeout estimation is a problem?  What's wrong with 
it -- it is setting timeout values too large?  That shouldn't cause any actual 
problems unless you were *also* having lots of transaction failures.

What operation is this -- download?  AFAIC think, download should almost *never* 
stop and wait for a timeout to elapse, so too-high timeouts should almost never 
cause an actual stall.

Now throttler is another story.  I suggest we nuke throttler from _new and then 
we'll put back a fixed one afterward (e.g. use Twisted's nice new throttler).

I suppose this is the tension between "make release" versus "clean up" desires.  
But whether your goal is a release or a cleaner code-base, I really don't think 
it is a good use of our time to tinker with the old throttling system.

--Z



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
mnet-devel mailing list
mnet-devel at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mnet-devel