Date: Mon, 6 Apr 1998 17:52:44 -0400 (EDT) From: Glen Foster <gfoster@gfoster.com> To: chrism@keyworld.net Cc: Anthony.Barlow@europe.simoco.com, freebsd-isp@FreeBSD.ORG, freebsd@plinet.com Subject: Re: Bandwidth limiter for services? Message-ID: <199804062152.RAA10319@gfoster.intr.net> In-Reply-To: <199804062104.XAA27968@mail.keyworld.net> (chrism@keyworld.net)
next in thread | previous in thread | raw e-mail | index | archive | help
Is this really the case? Dennis has been reluctant to describe the mechanism of his bandwidth manager but hints he has dropped on this and other FreeBSD lists indicate that they use a "strategically-timed ACK delay" mechanism to limit bandwidth rather than dropping packets. Basically, they ack as fast as the bandwidth limit allows given the size of the associated queue. Obviously, this adds a little or a lot of latency, more when the bandwidth limit is approaching, but this is not unlike the way a partially-meshed network "looks" to a sender as it approachs saturation and it conserves aggregate bandwidth better than an aggressive discard strategy does (at least as long as the ACK delay is shorter than the retransmission timeout). This is speculation on my part based on scanty evidence. I wish Dennis would 'fess up and describe the logic behind his code in this forum. Glen Foster <gfoster@gfoster.com> >From: "Christopher Martin at Home" <chrism@keyworld.net> >Date: Mon, 6 Apr 1998 23:08:02 +0200 > >I suggest that you check it out though. I think it works by dropping >packets. If it does the pipe might be filled by stuff that is already >downloaded. This might result in unaccepttable amount of retransmissions >from source. > >So you would not be really saving bandwidth... To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-isp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199804062152.RAA10319>