|
On Sat, 4 Nov 2000, Matthew S. Crocker wrote:
> On Fri, 3 Nov 2000, Horms wrote:
>
> > The key factors you need to take into account are:
> >
> > * The more packets/s you expect the more CPU power you will need.
>
> Would 2.4 be a better solution, doesn't it have better networking code?
> Would multi-processor help on the director?
only if you are CPU bound. Routing (VS-DR) doesn't take noticable CPU
power till you are running a Gbps network.
> > * Network bandwidth is usually a limitation.
>
> For web serving the inbound bandwidth is tiny compared to the outbound so
> a 100MB NIC should be plenty
you have to define tiny. My performance tests show that packets smaller
that the MTU take as long to transmit as a full sized packet. The ratio
which determines the throughput is the ratio of reply packets to request
packets. Still 100Mbps is easy for an LVS, but few people are going to
have a connection to the internet that size.
> Does anyone have any stats for this? CPUxRAMxNIC = x/connections/sec. I'm
> assuming a 133Mhz FSB would improve things a bit as well.
A 133MHz pentium I with 64M can handle 50Mbps in VS-NAT mode (see my
performance notes on the website).
Joe
--
Joseph Mack mack@xxxxxxxxxxx
|
