Date: Tue, 1 Jun 1999 12:40:34 -0700 (PDT) From: <kip@lyris.com> To: Nate Williams <nate@mt.sri.com> Cc: Poul-Henning Kamp <phk@FreeBSD.ORG>, current@FreeBSD.ORG Subject: Re: net.inet.tcp.always_keepalive on as default ? Message-ID: <Pine.SOL.4.05.9906011233100.5802-100000@luna> In-Reply-To: <199906011851.MAA14756@mt.sri.com>
next in thread | previous in thread | raw e-mail | index | archive | help
I think it is fair to say that the nature of the internet has changed somewhat since the standards were made. Keepalives by default are not sent until after two hours, if they are acknowledged no more packets are sent. If not 10 more probes are sent 75 seconds apart before the connection is declared dead. I think it somewhat silly to say that this is consuming a lot of bandwidth. The average mail message (4k) is 4 packets, the average telnet session is at least several hundred and an ftp session is going to be many, many more. Back in the day when people were arguing about the congestion it would create a 300baud modem was considered completely normal. Nowadays, when the average gaudy web page is > 20k (read ~20 1k packets) it is safe to say that things have changed. -Kip On Tue, 1 Jun 1999, Nate Williams wrote: > > Considering the number of hosts on the net today, which come and > > go with no warning and with dynamic IP assignments, I would propose > > that we disregard what the "old farts" felt about TCP keepalives, > > and enable the sysctl net.inet.tcp.always_keepalive as default. > > Seeing as the amount of traffic and congestion in the Internet, I > propose we diregard what the 'old fart' PHK says and not increase the > congestion with the use of keepalives. :) > > The 'old farts' did a good job of designing a system that happens to > work better than all of the systems the 'young farts' were able to > design. > > PHK's arguments are specious, since *any* traffic when the link is > congested is more congestion. > > > The argument against is that this will increas trafic and keep > > dynamic lines up when they should otherwise have been allowed to > > fall down. > > > > The former argument doesn't hold water, since we're talking about > > a TCP segment per hour (or less) per connection. > > That's still traffic, and congestion is congestion. On one systems that > isn't a lot, but with alot of connections it can add up to a significant > amount of bandwidth. > > > The second argument falls on the same reasoning in my book, I don't > > know of any on-demand lines with a timeout longer than 10 minutes > > anyway. > > You don't know of any, but that doesn't mean they don't exist. > > > Nate > > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-current" in the body of the message > > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SOL.4.05.9906011233100.5802-100000>