Date: Tue, 31 Jan 2023 08:39:44 -0700 From: David <2yt@gmx.com> To: freebsd-stable@freebsd.org Subject: Re: Slow WAN traffic to FreeBSD hosts but not to Linux hosts---how to debug/fix? Message-ID: <ee8df505-cd76-41fb-25a6-cb5c67dc5d8c@gmx.com> In-Reply-To: <CANwXMPNBkWQHzT4f%2Bjy0RV1gK7R6nJOWYsG5NZTf9RadRDNfXg@mail.gmail.com> References: <95EDCFCA-7E3F-458F-85A6-856D606B9D98@gromit.dlib.vt.edu> <tr9g44$d3l$1@ciao.gmane.io> <CANwXMPNBkWQHzT4f%2Bjy0RV1gK7R6nJOWYsG5NZTf9RadRDNfXg@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 1/30/23 16:30, Matt Garber wrote: > > > Any help/insight is gratefully appreciated. > > > > Cheers, > > > > Paul. > > > > sysctl net.inet.tcp.cc.algorithm=htcp > > I would set "htcp" on the server and home computer to improve > through in > your type of situation. > > > > There may be other FreeBSD sysctls that have bad defaults in this > scenario and could be better tuned, but I doubt changing the CC > algorithm at this point is the problem — at least not so much a problem > that’s causing throughput to be reduced so drastically. Happy to be > wrong if that does help things quickly and easily, though. > > (Since OP mentioned that FreeBSD CC was set to CUBIC, that would match > what the Linux boxes are using by default, too, unless they’ve been > changed to something newer like BBR… so seems like CUBIC *should* be > performing fine on this WAN link, and the difference is something else.) > > > —Matt > I love FreeBSD and very much appreciate the efforts of those people much smarter. But... I don't think the defaults get enough testing in real world conditions. I came across Paul's issue several years ago and spent a few a hours testing and found the defaults performed very well on a LAN but could perform terribly on a many hop WAN. HTCP performs marginally worse on a LAN or close WAN connection, but much much better on a many hop WAN connection. Testing: ------------------------------------ All computers are running FreeBSD 13.1. File transfer done with SCP. No changes were made between runs other than toggling the cc.algorithm. Be sure to run sysctl net.inet.tcp.hostcache.expire=1 before testing. (default 3600) HOST 1) 10 hops and 50ms pings. Connection limited to 120mb ------------------------------------ Run #1 NEWRENO 100% 315MB 13.0MB/s 00:24 HTCP 100% 315MB 12.8MB/s 00:24 Run #2 NEWRENO 100% 315MB 13.2MB/s 00:23 HTCP 100% 315MB 12.8MB/s 00:24 HOST 2) 17 hops and 110ms pings. Connection limited to 100mb ------------------------------------ Run #1 NEWRENO 100% 315MB 1.1MB/s 04:48 HTCP 100% 315MB 10.5MB/s 00:30 Run #2 NEWRENO 100% 315MB 1.6MB/s 03:22 HTCP 100% 315MB 3.5MB/s 01:29 Run #3 NEWRENO 100% 315MB 1.3MB/s 04:00 HTCP 100% 315MB 10.8MB/s 00:29 In my opinion HTCP is a better default for the current state of the internet. Cheers, David
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?ee8df505-cd76-41fb-25a6-cb5c67dc5d8c>