|
We have tested LVS with DNS, which is UDP based, too.
What we are doing with this test is not for heavy load
issue, rather to see if LVS can provide a fail-over
mechanism for the services. By load balancing the
servers, we can make two servers backup one, if the
one failed, the service will not stop.
If round-robin does this one request per server, it is
pretty hard to explain what we saw at the server log,
which indicating one server getting twice the requests
than other two in some seconds, and getting a lot less
at other seconds. Could you explain why we seeing
that?
Thanks!
At 06:40 PM 4/24/00 -0400, Joseph Mack wrote:
>On Mon, 24 Apr 2000, Wayne wrote:
>
> > I am testing using NTP.
>
>;)
>I was looking for something to test udp on and
>the best I could think of was ntp.
>
>Two issues: 1) round-robin
> > seems balance better than least connections; 2) the
> > balance seems in a large time frame is fine (by sampling
> > the NTP log every 5 minutes) but not fine by sampling
> > the NTP log every second. Is round robin sending
> > traffic to servers based on each request or based on
> > a period of time?
>
>each connection.
>
>I would expect least connection is meaningless in the udp context, since
>there aren't any connections ;) but quite what you'd see as far as
>balancing, I don't know.
>
>Are you doing this just for fun or are you really going to LVS ntp? I
>would expect ntp is well suited for distributed operation already and
>wouldn't need LVS.
>
>Joe
>
>--
>Joseph Mack mack@xxxxxxxxxxx
|
