|
> > In case, is it helpful to use Dual CPU or not?
>
> http://www.austintek.com/LVS/LVS-HOWTO/HOWTO/LVS-
> HOWTO.performance.html#FAQ:smp_doesnt_help
Yes it is true that for LVS it does not really matter, but LVS is not always
the only thing running on a system, in my case I also let it run a DNS
server and it has a lot of iptables rules. At the moment we are doing around
60 mbit/s and my stats are:
$ mpstat -P ALL 1 10
Average: CPU %user %nice %sys %iowait %irq %soft %steal
%idle intr/s
Average: all 0.23 0.00 0.69 0.13 0.48 2.93 0.00
95.54 11772.75
Average: 0 0.10 0.00 0.30 0.40 0.00 0.00 0.00
99.20 1000.30
Average: 1 0.30 0.00 1.01 0.10 0.00 0.00 0.00
98.59 0.00
Average: 2 0.10 0.00 0.41 0.00 0.62 1.24 0.00
97.62 2941.92
Average: 3 0.31 0.00 1.04 0.00 1.35 10.77 0.00
86.54 7830.64
(a quadcore xeon was just a little bit more expensive than a singlecore cpu
at the time we bought the servers).
Also you can have different cpu's handle the interrupts:
$ cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
1274: 220 4719154 2977300793 0 PCI-MSI-edge eth1
1275: 244 3649053 0 1213052367 PCI-MSI-edge eth0
With the speed of the current cpu's however I think it does not really
matter if you have a single or multi-core cpu, they all can handle gbit/s of
data.
>
> > and is it helpful to use 64bit x86_64 system or not?
> don't know. Presumably
It is helpful if you don't want stuff like /proc/net/dev overflowing (4
gbyte / 5 minutes = 16 mbit of traffic, so if you get above that, and you
try to use those statistics, the counter may overflow before you read it
again)
-kees
|
