Date: Fri, 7 Feb 2003 09:37:47 +0100 (CET) From: Harti Brandt <brandt@fokus.fraunhofer.de> To: Nate Williams <nate@yogotech.com> Cc: Doug Barton <DougB@FreeBSD.org>, chat@FreeBSD.org Subject: Re: High-latency/long-distance IP stack (was Re: cvs commit: CVSROOT access access.master access.ports) Message-ID: <20030207093019.O3730@beagle.fokus.gmd.de> In-Reply-To: <15939.36.855969.496240@emerger.yogotech.com> References: <20030205233916.6156C2A89E@canning.wemm.org> <20030205231630.D32815-100000@patrocles.silby.com> <20030206022854.G40993@12-234-22-23.pyvrag.nggov.pbz> <15939.36.855969.496240@emerger.yogotech.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 6 Feb 2003, Nate Williams wrote: NW>[ Moved to -chat ] NW> NW>> > Didn't Harti Brandt say that he was working on satelite communications? NW>> > Perhaps he has plans to allow even larger RTTs to work well. :) NW>> > NW>> > "FreeBSD: The best connection to Mars... or AOL" NW>> NW>> You laugh, but various people, including Vint Cerf, are currently working NW>> on an IP stack (or something that can be made to look like an IP stack) NW>> that will work across those kinds of distances. NW> NW>It really isn't that hard to do that, IMNSHO. The hardest part is NW>finding good 'initial' values for timeouts (if the first packet gets NW>lost), since you don't want *too* long if it happens that the link is NW>short. (Been there, done that. :) NW> NW>Once you get the initial 'broad' issues sorted out and understand that NW>packet loss != congestion, the above stack isn't that hard to NW>build/design. In fact there are such stacks: SCPS and now the CCSDS protocol suite. Some of the CCSDS protocols are designed to work over distances to mars (they are quite different from TCP, however). What is missing from current FreeBSD is clearly SACK. Even when doing TCP to the moon SACK can help. In the last two years I have seen at least two announcements of people that have offered a SACK implementation to FreeBSD. Unfortunately non of the TCP experts in the FreeBSD camp took a chance to get this into the tree. A minor problem is that there should be some documentation on how to tune FreeBSD TCP for long fat pipes. There are a number of non-abvious sysctls you have to tune to get it working. An yes, the current RTT stuff in FreeBSD is broken. I'm running my measurement machines on HZ=10000. RTTs are counted in ticks. There is a point where RTT computation breaks. harti -- harti brandt, http://www.fokus.gmd.de/research/cc/cats/employees/hartmut.brandt/private brandt@fokus.fraunhofer.de, harti@freebsd.org To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-chat" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030207093019.O3730>