Geoff,
>You are of course correct indicating that the servers need to operate
>with a TCP stack which operates with a larger initial cwnd window size.
>
>What's the incentive to make the change on a particular web server?
>
> The web server looks a damn sight faster than others, particularly
> for short transactions.
Absolutely - unless it turns out that it makes things *worse* for a busy
web server (or it's downstream path); At this point, it's still a
risky decision.
Personally, I'm anxious to see one or more busy web sites (with extensive
performance monitoring) actually try using an increased I-CWND; Real-life
experience and measurements should be the driver in determining any default
parameter settings.
If I were a vendor (I'm not) selling drop-in web-server boxes for use by
mere mortals, then I'd want to be darned sure that my shipped defaults
give the best general performance. If a buyer wants to tweak from there,
that's cool. Performance tuning is what consultants live for... :o)
I really don't want to start a sub-thread on I-CWND (the draft went into
final call; this isn't the place to discuss the pros/cons), but in the
context of what a vendor should ship in as default settings - my
non-scholarly interpretation of the discussion regarding the (optional)
increase of I-CWND :
In all likelihood, it probably will NOT melt the Internet backbone, but
there are also no guarantees that it won't degrade things in the first
hop or so in the end-to-end path from a busy server. Since any resulting
degradation will be mostly self-imposed) (the entity with the *busy*
server probably "owns" the first hop or two toward the backbone), allowing
an *optional* increase in I-CWND is probably OK.
Taken in terms of
"Be conservative in what you send, liberal in what you receive"
One should only increase their default I-CWND setting only after it has
been determining that it will not degrade their end-to-end performance.
I-CWND just becomes another parameter available for performance tuning.
We are given enough rope to hang ourselves, where we go from there is
up to us.
To me, all of this means that vendors should ship with conservative
default settings (on everything!) and let the buyer/user/administrator
configure things for better performance.
>In my humble view is IS a large incentive, particularly when the
>perception of most of the web as a user experience is that it is
>just too damn slow!
I kind of blame that on the content - too many flashy pictures and
advertising banners, very little text. I've come to love the speed
and simplicity of using a text-based browser whenever possible. It
really speeds things up. I realize that I'm in the minority on this
one :o)
And with that, we've managed to totally leave the domain of this
mailing list - sorry!
Regards,
Eric
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:43 EST