Date: Mon, 10 Apr 2006 15:29:02 -0400 From: "Zaphod Beeblebrox" <zbeeble@gmail.com> To: "Peter Jeremy" <peterjeremy@optushome.com.au> Cc: freebsd-stable@freebsd.org Subject: Re: Disappointed-new Message-ID: <5f67a8c40604101229m73e5afa0n619a6725a7168e46@mail.gmail.com> In-Reply-To: <20060410080418.GB739@turion.vk2pj.dyndns.org> References: <20060407121452.GO1784@math.jussieu.fr> <6.2.3.4.0.20060408150025.099369a8@64.7.153.2> <5f67a8c40604091456gfef47d3q3583d3d1a519d035@mail.gmail.com> <20060410080418.GB739@turion.vk2pj.dyndns.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 4/10/06, Peter Jeremy <peterjeremy@optushome.com.au> wrote: <metoo> > I can't comment on gigabit performance but at Fast Ethernet speeds, > I've found that fxp performs much better than dc, tl and tx NICs. > I've had fairly bad experiences with bge under even moderate load > (though the one in my laptop seems OK). > </metoo> > One problem everyone might be facing is that both BGE and EM are good examples of bad vendor practice --- specifically making significant changes to a product without making it clear set of features you have/don't have. FreeBSD, in a way, makes this work by calling a whole class of cards by it'= s driver name. Both EM and BGE have versions that omit important features. My old laptop, for instance, had a BGE that could not send large frames. Many of the interrupt coalescing functions that are so amazing in EM are options that are deleted from cheaper versions of the chipset. Even the lowly FXP comes in more than a dozen flavours (although most of th= e differences ... like on-chip encryption ... are not supported in *BSD). I can honestly say that while BGE gave me great pains when they first came out, of late I havn't had cause to complain. Caution (based on past experience) prevents me from using them in servers, not current ongoing testing. Nothing I've read has given me any indication that they are capable of performing like good EM's (we only buy the server product, not the workstation version). Now... keep in mind that our stress tests care as much about 64 byte packet performance as they do large packet throughput. Average router pakcet size is on the way down, not up. Even if you estimate 350/550 as your average (out/in packet sizes), you care about 64 byte performance because of DDOS.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5f67a8c40604101229m73e5afa0n619a6725a7168e46>