oskar@is.co.za writes:
>Hi
>
>Someone has been trying to get us to by a 'Shiva PowerSurf' box...
>
>It's "patent pending" (which means their techies will say "It's patent
>pending" whenever they don't know or don't care, it seems)
>
>It seems that when you connect to a site it either scans the contents
>of the HTML to get a list of all links on the page and then opens
>TCP connections to each of them, or it simply connects to the same
>site with an 'empty pipe' just in case the next graphic/connection goes
>to the same site.
>
>This supposedly increases BW usage by 4%, but reduces latency dramatically.
>
>I don't suggest that we do the same thing in squid, but what about
>having an option so that when you setup a parent it opens a few connections
>to the TCP port 'for future connections' to reduce the latency? Yes I know
>'persistent connections' would fix this, but if you have a parent
>that doesn't support this it might be useful. Eventually we could
>change the number of 'threads' to the remote server depending on load.
I actually implemented this once to see if it made any difference. I
couldn't detect any improvement within the bounds of uncertainties, but
that may have been my poor methodology.
Anyway, it was pretty easy to do.
BUT.... IMHO, I really think we should rely on persistent connections.
I guess I'd like to avoid adding another no-so-often used feature
to Squid just because we can...
Duane W.
Received on Tue Jul 29 2003 - 13:15:44 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:27 MST