[lvs-users] Issues with weighted round robin

Himanshu Raina himanshu.raina at naukri.com
Thu Jul 29 11:59:40 BST 2010

The current stats output is as follows

Prot LocalAddress:Port               Conns   InPkts  OutPkts  InBytes

TCP               1877189    39344549       0   11215M
  ->               599836     25601414        0    7877M
  ->              1277353      13743135      0    3338M

On Thu, 2010-07-29 at 16:20 +0530, Graeme Fowler wrote:

> Hi
> Firstly, your reply to the list didn't go through because somehow you
> mangled the address.
> On Thu, 2010-07-29 at 16:09 +0530, Himanshu Raina wrote:
> > I understood the connection establishment and the termination part. So
> > one way of dealing with this is to assign weights accordingly. Knowing
> > that the server with higher weight processes request much faster you
> > should assign maximum weight to it. Like in my case when I changed the
> > weight from 7:3 to 9:1 the number of active connections reduced on the
> > server which was earlier having higher number of active connections and
> > every new connection in this case was given to the server with higher
> > weight. Correct me if I'm wrong.
> Please supply the --stats output as requested after running for a while.
> The "snapshot" view of active/inactive connections is not very helpful
> in this discussion.
> The faster server can move more connections more quickly from active to
> inactive than the slower, so can appear to be handling fewer connections
> when viewed in this way. You need to look at the "big picture" after
> it's been running for a while to get a true view of how requests are
> being spread between realservers.
> Graeme


Himanshu Raina 
A great empire, like a great cake, is most easily diminished at the edges.
		-- B. Franklin

More information about the lvs-users mailing list