LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: [LVS - NAT] alternatives

To: Nicolas Chiappero <Nicolas.Chiappero@xxxxxxxxx>
Subject: Re: [LVS - NAT] alternatives
Cc: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
From: Julian Anastasov <ja@xxxxxx>
Date: Tue, 11 Sep 2001 23:16:21 +0000 (GMT)
        Hello,

On Tue, 11 Sep 2001, Nicolas Chiappero wrote:

> 2. Playing with VS-NAT, I ran some tests using testlvs
> and I was wondering if I could find others results to
> make some performance comparisons...

        testlvs can't be used to measure the output NAT traffic,
so you have to use other applications for this. But with testlvs
you can analyze the throughput in the out->in direction on
different request rates and number of clients. I'm sure if you make
some tests with different number of clients you will see that the
problem for this small number of packet rate is not in the LVS. 2.4
behaves better.

> ./testlvs VIP:80 -tcp -srcnum 100000 -packets 5000000


> 'vmstat 1' while running test :
> r  b  w   swpd   free   buff  cache  si  so    bi    bo    in    cs  us  sy
> id
> 0  0  0      0 159936  18656  16260   0   0     0     0 19204    13   0   1

        This is no more than 19204 packets/sec. BTW, my tests show
similar rates when I'm using a plain 2.2 kernel with an low-speed
motherboard and that the 2.2 measurement for the CPU times is not
very accurate on packet flood.

> * RealServer *
> ./show_traffic.sh eth0 : 3150 packets/s

        This is expected, you are using too large number of clients.
Your Linux 2.2 is not tuned for this. And you are flooding only with small
SYN packets. You can try with 100-byte UDP packets too.

> Regards,
> Nico.


Regards

--
Julian Anastasov <ja@xxxxxx>



<Prev in Thread] Current Thread [Next in Thread>