[lvs-users] Client, Server and Director in same LAN

[Malte Geierhos]

Hi, 

Am 30.10.2010 um 09:35 schrieb Michael Schwartzkopff:

> On Saturday 30 October 2010 07:10:17 Simon Horman wrote:
>> On Sat, Oct 30, 2010 at 05:54:57AM +0200, Michael Schwartzkopff wrote:
>>> On Saturday 30 October 2010 05:03:33 Simon Horman wrote:
>>>> On Thu, Oct 28, 2010 at 03:59:17PM +0200, Michael Schwartzkopff wrote:
>>>>> On Thursday 28 October 2010 13:42:47 Simon Horman wrote:
>>>>>> On Thu, Oct 28, 2010 at 08:24:36AM +0200, Michael Schwartzkopff 
> wrote:
>>>>>>> Hi,
>>>>>>> 
>>>>>>> is it possible to use direct routing with clients, real server
>>>>>>> and director in the same LAN?
>>>>>> 
>>>>>> Yes. Actually, that is how I do most of my testing.
>>>>> 
>>>>> Do you have any idea why the director might limit the performance of
>>>>> our system? Without LB I get 100 connections/s to the real IP
>>>>> address of a real server. When I address the virtual IP of the
>>>>> director I get a performance drop to 1 connection/s.
>>>>> 
>>>>> tcpdump shows that sometimes I have no traffic at all on the line to
>>>>> 0.2 sec.
>>>>> 
>>>>> Any ideas? Thanks.
>>>> 
>>>> If you are running 2.6.36 then this may relate to a performance
>>>> regression related to the introduction of double NAT.
>>>> 
>>>> But regardless, that is a pretty startling result that
>>>> I don't have a decent explanation for.
>>> 
>>> No. have some older version on the kernel and do not use NAT at all.
>>> 
>>> The only explanation we found at the moment is that something with
>>> bonding did not work properly.
>> 
>> Would it be possible for you to try this with a newer kernel
>> and if pain persists describe your setup in a little more detail?
> 
> We have 2.6.34.
> Simple setup. Everything in the same LAN. direct routing. All hosts with 
> bonding interfaces connected to two switches.

And you're quit sure that every host is "active" on the same switch ? 
I mean i guess you're aware of a non-existing broadcast domain - across those switches 
and that in mind - perhaps problems with the layer2 rewriting ... the director does ?

i'd suggest testing first without bonding with only one switch - and then after all 
configure the bonding in a failover scenario - i don't know about your switches - but 
i'm guessing you're not using these 10k+ € pieces, that can do LACP over several switches ..

regards, 
Malte

> 
> Greetings,
> 
> -- 
> Dr. Michael Schwartzkopff
> Guardinistr. 63
> 81375 München
> 
> Tel: (0163) 172 50 98
> _______________________________________________
> Please read the documentation before posting - it's available at:
> http://www.linuxvirtualserver.org/
> 
> LinuxVirtualServer.org mailing list - lvs-users at LinuxVirtualServer.org
> Send requests to lvs-users-request at LinuxVirtualServer.org
> or go to http://lists.graemef.net/mailman/listinfo/lvs-users

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4227 bytes
Desc: not available
Url : http://lists.graemef.net/pipermail/lvs-users/attachments/20101105/d5fd7c50/attachment-0001.bin