Eric,
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.
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!
geoff
At 23:57 7/05/98 -0400, Eric Travis wrote:
>George,
>
>: In a similar vein, can anybody comment on the likelyhood of a general release
>: of code which has a larger initial windowsize in PC platforms? I've heard
>: an interesting rumour to that effect here, which I'd love to see confirmed.
>:
>: This is in the context of non-tcpsat usage: simple changes which might improve
>: end-to-end performance for small-object web transfers, and time to achieve
>: full window state on the network in general.
>
>OK, I'll admit that I'm shooting from the hip here but -
>
>For any increase in the default initial congestion window of a connection
>to have a performance impact (positive or negative), they'll need to be
>enabled at a connection's data source.
>
>For things like http traffic, this means that as long as web-servers have
>an initial cwnd of 2 or more segments, then the client stack needs *no*
>modification at all of its default parameters. Further, since most people
>are sucking data from servers rather than sourcing it, there is no need
>to tweak *this* parameter in most home (or office) PCs.
>
>On the other-hand, this means that "J. D. User" needs every random
>web-server to be configured with a larger than 1 initial cwnd. This
>might happen, but it is going to be beyond the influence of individual
>client machines.
>
>Further, if someone is running a server where they need to be concerned
>about performance (I-CWND increase is a performance tweak), then they
>should be *expected* to do at least minimal performance tuning (including
>setting the default I-CWND) or they should seriously consider paying a
>commercial provider to host their web-site.
>
>In any case, there doesn't seem to be a compelling reason for a vendor
>to have the default I-CWND set to anything but one segment upon OS
>install. A knob allowing change from the default of one segment should
>be sufficient.
>
>: I think the days when people would volountarily load an alternate WINSOCK.DLL
>: like trumpet to get some outcome have gone. It has to come bundled on the CD
>: and enabled by default, or auto-enabled when needed to be useful in a dumbed-
>: down world...
>
>I don't know, even the least technical of users have gotten *more* used
>to (have come to expect the need to?) the need to reinstall the OS, or
>apply even to service patches.
>
>However, since it seems that the latest winsock specs allow for multiple
>transport providers to coexist, it might be as simple as littering a
>hard-disk with another TCP implementation at the same time that a driver
>is being installed and auto-magically doing surgery on the system registry.
>Actually, if all that is required is a registry tweaking, then the
>situation gets even simpler (the install procedure for a new browser
>can handle this for you).
>
>I need to be somewhat optimistic about the possibility of dropping in
>an alternative DLL for a new protocol stack since we're attempting to
>do something very similar for use with Windows-NT. :o(
>
>Regards,
>
>Eric
>
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:42 EST