On (2010-03-19 10:20 -0700), Stephen Cobb wrote: > Found what I was talking about...this may explain things: > https://puck.nether.net/pipermail/cisco-nsp/2007-April/039999.html
Just to throw in some speculation, as people probably know NPE-G1 is MIPS and NPE-G2 is PowerPC, so rather large change. Software routers tend to do context switch between every packet at low volume, to minimize latencies, so copy packet, lookup, copy packet and so forth, when CPU is starting to be congested, you copy more the one packet before looking them up, increasing latency slightly by scaling further. Maybe the new PowerPC architecture has more expensive context switch and performance might be rather linear with NPE-G1 until you're doing fewer context switches. It would be interesting to see how latencies in NPE-G1 and NPE-G2 differ at 100, 200, 400, 800Mbps. -- ++ytti _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
