You can take a look at the NAPI patches on ftp://robur.slu.se/pub/Linux/net-development/NAPI/ This includes changes for e1000 and tg3 also ... Thanks Manish --- Mathias Gug <Mathias.Gug@cern.ch> wrote: > Hello, > > I've performed some benchmarks using 2.4.19/20 > kernels and the tigon3 > driver. I've found that 2.4.19 performs better than > 2.4.20. > > > > > > This is the hardware that I've used : > > ** host 103 and host 104 : > > Dual Pentium 3 1 GHz - 512 RAM > Network Card : 3c996-T > (identified in /proc/pci by : > Ethernet controller: BROADCOM Corporation NetXtreme > BCM5700 Gigabit > Ethernet (rev 18).) > > redhat 7.2 > > ** host 110 : > Sinlge Pentium 4 2.40 GHz - 256 RAM > Network Card : 3c996-T > > redhat 7.2 > > ** network switch : > 3com 17700 - GE network > > > > > > I've used two benchmarks : > > ** lmbench-3.0-a2 : > > on the server : > # ./bw_tcp -s > > on the client : > # ./bw_tcp target_ip > > ** iperf version 1.6.4 (3 Dec 2002) pthreads > > on the server : > # ./iperf -s > > on the client : > # ./iperf -c target_ip -t 60 > > > > > > > I've used the following monitor : > > ** sar from sysstat-4.0.1-2 > on every node : > sar -I irq_network_card -n DEV -u 3 0 > > > > > > Here are the results (running every benchmark 10 > times and taking the > average results) : > > ** bw_tcp : > > server 103 - 2.4.19smp > > clients 104-2.4.19smp 104-2.4.20smp 110-2.4.19up > 110-2.4.20up > > bw 101.9 63.7 89.2 76.3 > (MB/s) > > intr/s (reported by sar) > client 6743 40165 6772 39680 > server 5628 4018 6705 6674 > > ** iperf : > > client 104 - 2.4.19smp > > servers 103-2.4.19smp 103-2.4.20smp 110-2.4.19up > 110-2.4.20up > > iperf 116.6 96.5 107.2 79.9 > (MB/s) > > intr/s (reported by sar) > client 6775 6799 6727 6774 > server 6800 49029 6772 39515 > > > > > > As you can see there is a loss in performance > between 2.4.19 and 2.4.20 > and a big rise in the number of interrupts > generated by the network card. > > So here are my questions : > > 1) Is NAPI enabled in 2.4.20 ? If not, how could I > enable it ? The idea > behind these tests is to evaluate NAPI. > > 2) Why is there such a high rate of interrupts > generated by the network > card ? > > > Thank you > > Mathias Gug > > - > : send the line > "unsubscribe linux-net" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html __________________________________________________ Do you Yahoo!? Yahoo! Mail Plus - Powerful. Affordable. Sign up now. http://mailplus.yahoo.com - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
