|
On Sun, 11 Jun 2000, Vivek Raghunathan wrote:
> My machines are connected via a 10 MBPS Ethernet. With my LVS up and
> running, my performance drops (yes!) to about 75% of the value for a
> single machine.
> using are all Linux Kernel 2.2.12-20 machines with apache 1.3.
> P-III/550MHZ + 128 MB RAM + Realtek 8029 (PCI) cards (pci-ne2k.c)
This sort of machine should be able to give 400Mbps to the tcpip stack,
and 100Mbps to a 100Mbps ethernet. If you have 10Mbps ethernet with coax
or a hub, throughput will drop catastrophically from collisions at about
60-80% capacity.
Do you want to measure the latency added by LVS? If so you'll need a
network with a lower latency than the LVS. You'll proably need to go to
expensive networking, like myrinet with the CPU speed you have or else you
could lower the speed of the CPU, testing on a 33MHz machine. This would
be like putting LVS into slow motion.
Do you want to measure the performance of a particular setup of LVS? You
have already measured the properties of an LVS operating on a saturated
10Mbps ethernet. 100Mbps networking is now as cheap as 10Mbps was about 4
years ago. A typical LVS will now be running on a 100Mbps network. You
could rerun your tests with 100Mbps NICs and a 100Mbps switch.
> COuld you please
> 1. suggest a better traffic generator than Apache Bench..
still not a solved problem.
Joe
--
Joseph Mack mack@xxxxxxxxxxx
|
