[lvs-users] Issues with weighted round robin

[Graeme Fowler]

On Thu, 2010-07-29 at 15:11 +0530, Himanshu Raina wrote:
> The correct output of " ipvsadm -L -n " is given below. There was
> mistake in the earlier output.

Can you please give us the output of:

ipvsadm -Ln --stats
ipvsadm -Ln --rate

In your context, these are more meaningful. "-Ln" by itself is a
point-in-time view, "--stats" gives figures which are cumulative.

Just so you're aware; the WRR scheduler assigns *new* jobs according to
weight. The server with the lower spec cannot process them from SYN to
SYN/ACK to ACK/PSH to RST and FIN as quickly as the other box, so at any
time may appear to have more "active" connections. Over time, however,
you should find that the number of requests are spread between the two
in the ratio you define in their weights.

Graeme