From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ben Greear Subject: VIA Rhine (VT6105M) v/s Tulip performance. Date: Mon, 14 Mar 2005 19:57:15 -0800 Message-ID: <42365D1B.2080504@candelatech.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit To: "'netdev@oss.sgi.com'" Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org Hello! I just got a new 4-port VIA VT6105M NIC (router-board 44) and decided to benchmark it against my 4-port tulip NIC (p430tx)... My test is pktgen sending to itself across two interfaces, and my network emulator bridging the other two interfaces. It's wired so that pktgen is sending a packet through the bridge. There is no protocol overhead here since my bridge software doesn't care, and pktgen is receiving before the UDP protocol can get a hold of it. Kernel is 2.4.29, P-IV 2.66Ghz, 512MB RAM, RH9 w/X and some applications running. With the VIA board, I get about 39Mbps tx + rx on each interface with 1514 byte packets. I get a few dropped packets, maybe .1% or so. The tulip board sustains about 58Mbps tx+rx on each interface, and for whatever reason, the CPU usage is lower. I am dropping a larger percentage of packets with tulip, but not too bad. Both NICs are stable under load so far...though I've only cooked them for 10 minutes at a time. Will let the VIA cook over-night to see if it holds up (have cooked the tulip NICs for days and they always do well)... The VIA board is significantly cheaper, so the lower performance may just be the reality of the situation. I'll try the VIA in a 2.6 kernel sometime soon and let you all know if that changes things significantly. Take it easy, Ben -- Ben Greear Candela Technologies Inc http://www.candelatech.com