On Mon, 28 Aug 2000, Pietro Ravasio wrote:
> I do know that testlvs is not realized to test realserver performances, but
> I've got a little problem while running testlvs: each time I launch a test
> (even if with a low number of sent packets, ie. 10000) apache on real
> servers start to not to respond anymore (even if it seems to be running (ps
> -aux |grep httpd, but if I try a connection to localhost on one of the
> realserver I can't get anything). The only solutions is to restart apache.
This result is expected. Have you seen DDoS :)
> Has anyone of you experienced such a problem?
You are brave people :)
This tool is for people that often ask "Hm, what CPU I need for
LVS? Can LVS supprt N connections/sec? Is 128MB RAM enough for LVS?".
But if you want to hit your real servers, go ahead :)
> I'm running apache on two real servers with two NIC interfaces on each of
> them (and so 4 real server in LVS point of view), both of them configured
> to route responces via the same NIC interface requests came from (ip ru +
> ip ro... )
The packets sent from testlvs can't be used to complete a
connection setup. The source IP you use does not know anything about
the request sent from testlvs. You don't expect connection establishment,
right? I hope you don't send traffic outside, without the reject rule :)
> P.S.: I do experience this problem even if I add a "reject rule" in routing
> table for testlvs requests (route add -net 10.0.0.0 netmask 255.0.0.0 reject).
With this rule used the impact is only for the routing cache
but that can't be classified as problem:
cat /proc/net/rt_cache | wc -l
Julian Anastasov <ja@xxxxxx>