Date: Wed, 13 Aug 2003 14:26:15 -0400 From: Nick Evans <nevans@talkpoint.com> To: freebsd-performance@freebsd.org Subject: Re: Tuning Gigabit Message-ID: <20030813142615.084b383d.nevans@talkpoint.com> In-Reply-To: <16132.36223.175148.275860@canoe.velocet.net> References: <20030628190036.0E06B37B405@hub.freebsd.org> <000f01c33dad$1595a0f0$e602a8c0@flatline> <16126.9805.829406.368426@canoe.velocet.net> <000901c33dd1$12268780$0200000a@fireball> <16126.19861.842507.318997@canoe.velocet.net> <001f01c33e0f$1f4716d0$0200000a@fireball> <16127.2826.306427.946086@canoe.velocet.net> <3EFFC3D5.7070605@mullet.se> <3F003092.30405@centtech.com> <16128.42007.75314.462685@canoe.velocet.net> <3F043F3A.1030907@centtech.com> <16132.36223.175148.275860@canoe.velocet.net>
next in thread | previous in thread | raw e-mail | index | archive | help
I'm currently forwarding 700 megabits (1500 byte packets, netperf generated) , ~95K pps, with 254 ipfilter rules. The system is running at ~30% idle. I'm also running freevrrpd so the two units I have are fault tolerant with a 3 second failover time. Still tuning, more to come later. Nick On Thu, 3 Jul 2003 16:09:35 -0400 David Gilbert <dgilbert@velocet.ca> wrote: > >>>>> "Eric" == Eric Anderson <anderson@centtech.com> writes: > > Eric> What would be a good initial start hardware-wise? > > To start testing this type of setup (gigabit routing), you'd need a > vlan'able switch with many gigabit ports, and then for each test you > need: > > 1 or more packet generators (one interface each) > > 1 router (two interfaces) > > 1 or more victims (one interface each) > > For bidirectional packet passing tests, it's best to have two (or > more) generators and 2 or more victims. The victims generally have > the least load and are good places to measure actual traffic passed. > > It's a good idea to have serial (or other) console on each of these > machines as livelock is not uncommon. There are also useful > statistics you can collect from the console that you can't always get > from the network login ... because the near-livelock conditions may > not give you enough updates. > > It is also sometimes a good idea to have a 100meg card in the boxes > for OOB management. > > Multiply this times the number of tests you want to do. We find that > you get about one test every two hours total work (much tearing down > and building up of machines inbetween included in that estimate). > > Dave. > > -- > ============================================================================ > |David Gilbert, Velocet Communications. | Two things can only be | > |Mail: dgilbert@velocet.net | equal if and only if they | > |http://daveg.ca | are precisely opposite. | > =========================================================GLO================ > _______________________________________________ > freebsd-performance@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-performance > To unsubscribe, send any mail to "freebsd-performance-unsubscribe@freebsd.org"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030813142615.084b383d.nevans>