Date: Thu, 12 Apr 2007 20:36:13 +1200 From: Andrew Thompson <thompsa@freebsd.org> To: Adrian Chadd <adrian@freebsd.org> Cc: Ian FREISLICH <ianf@clue.co.za>, freebsd-current@freebsd.org Subject: Re: CFT: new trunk(4) Message-ID: <20070412083613.GG9390@heff.fud.org.nz> In-Reply-To: <d763ac660704120059ke89fdf5la750d95eafe49a41@mail.gmail.com> References: <peterjeremy@optushome.com.au> <20070411191450.GE815@turion.vk2pj.dyndns.org> <E1Hbs1M-000FWA-7Z@clue.co.za> <d763ac660704120059ke89fdf5la750d95eafe49a41@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Apr 12, 2007 at 03:59:53PM +0800, Adrian Chadd wrote: > On 12/04/07, Ian FREISLICH <ianf@clue.co.za> wrote: > > >This is the configuration I'm going to be testing with our provider > >this morning because FreeBSD can't do link aggregation on VLAN > >interfaces. I'm hoping that not having IP data available to the > >switch will not prevent it from working in our scenario. > > Some switches (I'm sure the Cisco ones, for example) will happily do > etherchannel packet distribution based on IP header info even if > they're just configured to switch and not route. > > All it has to "know" about IP is which offsets in the packet > correspond to what it should be hashing against. And (cisco) switches > have been blurring the L2/L3 switch boundary for a while (think IGMP > snooping for a very early example.) In fec, lacp and loadbalance modes our trunk(4) also distributes using a hash function, it hashes the vlan tag (if present), src and dst mac addresses, and the full ip/ipv6 header (if present). See the trunk_hashmbuf() function for the implementation. The only trunk mode that isnt safe for frame ordering is roundrobin, and just alternates ports as the name suggests. OpenBSD has this as the default which is quite suprising, I changed our default to failover. Andrew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070412083613.GG9390>