Problem accessing to Windows Terminal Server in load balancing.

Problem accessing to Windows Terminal Server in load balancing.

[ArcosCom Linux User]

Hi guys, I have this configuration:


LAN -- LINUX BRIDGE -- LINUX BRIDGE -- TS

The linux's bridges are for wifi purposes (2 centos machines) and is
working fine.

The problem I have is that the TS are 3 windows TS in load balancing
configuration. There is 1 IP for general purposes and every TS has its own
IP.
For example, the configuration is:

    192.168.0.1/24  TS1
    192.168.0.2/24  TS2
    192.168.0.3/24  TS3
    192.168.0.10/24 (TS server IP for load balancing)

The problem I encountered is that every PC into the LAN has access from
the LAN to every TS (tested with ping), but them have no access to the IP
for load balancing.

I observed that all TSn IP's has their ARP entry in the ARP table, but the
IP for load balancing not.

How can I solve this problem at the bridge level? I think that there is
something that the bridges are filtering and perhaps is because the TS
load balancing software uses various "virtual macs" for the load balancing
IP.

Perhaps I must change any /proc/sys/net/ipv4/conf/<dev>/ file to bypass
this filter, I'm thinking in "arp_ignore", put a "1" into it, but is a
production environment and I don't want to loss the access to the devices.

Any help?

Re: Problem accessing to Windows Terminal Server in load balancing.

[Grant Taylor]

On 05/28/08 03:54, ArcosCom Linux User wrote:
> Any help?

Please provide more information on how you are doing your "Load Balancing".

The problem you initially described sounds a lot like "The ARP Problem" 
with the real servers being on the same subnet as the client machines.



Grant. . . .

Re: Problem accessing to Windows Terminal Server in load balancing.

[Thomas Jacob]

I'm guessing you're using Windows Network Load balancein (NLB)?

NLB relies on all incoming traffic being sent to all the nodes,
the way it does this is by either not sending outgoing ethernet
frames from the cluster MAC address (unicast mode, so that ordinary
switches don't learn its origin and broadcast all the data) or by using
a multicast ethernet address (multicast mode) as the cluster MAC. Either
way you are flooding you L2 broadcast domain with all incoming traffic, 
this is not always a clever idea.

Maybe this creates problems for Linux bridging?

     Regards,
              Thomas