> From: George Michaelson <[email protected]>
>
> I suppose if the median transfer size was smaller I'd be on stronger ground
> but not much. As others have pointed out, server side is what matters.
Interesting point - this might be solvable by a MSS-style option:
send SYN + "request for larger initial windiw"
routers on the path stomp on the "request..." if their queues are
backlogged (RED-style, maybe even)
this is the analog of stepping on the MSS request
server decides what to use as an IW
However - the problem is that the queues on the return path are what are
important, and paths are not required to be symmetric (and aren't, esp.
in the satellite case). The servers would need ongoing state to know
whether to use a larger IW for new connections (see RFC 2140 for a description
of how this might be used). This state can be shared among hosts on
a LAN, regarding similar remote paths (also in RFC2140).
> And both persistent connections and multiple connections defeat this.
>
> How so? multiple I can understand. but persistance has to win surely?
Persistence means the IW is used only once, when the connection opens,
and not for subsequent requests. That uses the restart window size,
which is more likely to be tied to the loss window size (see Partridge's
description of these three parameters).
Joe
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:43 EST