LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: IPVS Benchmarking

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Re: IPVS Benchmarking
From: Horms <horms@xxxxxxxxxxxx>
Date: Sat, 15 Jan 2000 13:34:33 -0800
On Wed, Jan 12, 2000 at 11:45:03PM +0800, Wensong Zhang wrote:
> Hi Horms,
> 
> I don't know if your node-3 is a linux box with two ee1000 cards or
> gigabit switcher/router. If your node-3 is a Linux box, I think that
> the above network topology is good for benchmarking LVS/NAT and the
> node-3 take the IPVS job. because ee1000 cards on Linux has
> performance issues.
> 
> For LVS/DR or LVS/TUN, you might change your network topology like
> 
>        node-1      node-6     node-7
>        (client)   (client)   (client)
>            |         |          |              gigabit network
>   ---------+---------+----------+---------+--  192.168.1.0/24
>            |         |          |         |
>          node-2    node-3     node-4    node-5
>          (IPVS)   (server)   (server)  (server)
> 
> All the clients and servers are in the same network. Then, the maximum
> throught 185Mbits/s/node won't affect benchmarking the maximum
> throughput of the whole server cluster earlier.

I have tested a topology like that with 100Mbit cards and
the effective throughput for n clients and n servers
was n*100Mbit, as ICMP redirects were issued by the IPVS box.
While this is great for wow number it doesn't really show
much because clients generally aren't going to be on
the same LAN as servers.

-- 
Horms

----------------------------------------------------------------------
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
To unsubscribe, e-mail: lvs-users-unsubscribe@xxxxxxxxxxxxxxxxxxxxxx
For additional commands, e-mail: lvs-users-help@xxxxxxxxxxxxxxxxxxxxxx

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