On Tue, 2004-04-27 at 15:08, John Reuning wrote:
> Does anyone know where I could look to find and fix a performance
> ceiling with an LVS director that can't handle > 200 syn pps? I enabled
> irq affinity and napi (dual e1000 adapters), but it only raised the
> limit from ~150 syn pps to ~200 syn pps.
Are you sure the bottleneck is in LVS, and not just in Linux as a packet
I did some kernel profiling and found that iptables and ipvs accounted
for only 20% of the work that a dual p4(htt) director was performing at
its MLFFR (Maximum Loss Free Forwarding Rate). The MLFFR for straight
forwarding was 266Kpps, while for LVS-DR w/ fwmarks it was 230Kpps.
So I'd look elsewhere for the bottleneck.