[Spread-users] multihomed configuration problems

[Jonathan Stanton]

Hi,

I noticed one or two specific issues that may be the cause of your 
problem with the multi-home support. I am also curious what aspects of 
Spread's performance you want to improve by moving it to a separate 
network segment? In many wackamole setups, the spread traffic is minimal 
so throughput isn't usually an issue. Can you tell us if you saw a 
particular performance problem or if you were trying to prevent future 
problems :-)

On Mon, Dec 06, 2004 at 12:38:51PM -0600, Eric L. Anderson wrote:
> I have successfully managed getting both spread and wackamole working
> with wackamole managing some IP's on a public network using two
> machines. I thought I might improve the performance of spread by
> moving the spread communication to a private network using
> mutltihommed machines while still having wackamole manage the public
> interfaces. I used the multihomed example from the spread.conf but
> things are not working as I expected within spread.
>  
> The machines in question have dual Broadcom adapters running FreeBSD
> 5.3 and spread 3.17.2. For some reason, all I ever see in the logs is
> the following (the public IPs have been changed to protect the
> innocent) :
>  
> # tail /var/log/spread.log
> [Mon 06 Dec 2004 11:29:06] Finished configuration file.
> [Mon 06 Dec 2004 11:29:06] Conf_init: My name: host1, id: 1.2.3.7, port: 4803

If nothing else is printed then for some reason they are not finding 
each other. You should usually see a set of lines like:

 Configuration at lothlorien is:
 Num Segments 1
     1       10.0.1.255        4803
             lothlorien              10.0.1.107      
 ====================

listing the other daemons that were found.

> 
> And neither of the machines are communicating to each other over (even
> though the 192.186.0  IPs are pingable from each machine.
> 
> Here is the spread.conf for both machines:
> 
> Spread_Segment 225.0.1.1 {
> 
>         host1     1.2.3.7 {
>                 D               192.168.0.2
>                 C               1.2.3.7
>         }
>         host2       1.2.3.4 {
>                 D               192.168.0.1
>                 C               1.2.3.4
>         }
> }

I'm guessing you are running linux. Linux and IP Multicast have a number 
of quirks and outright annoying behaviours. 

As a quick test you might try switching the 225.0.0.1 to the broadcast 
address for 192.168 and see if that works. If so, then you probably just 
have to fix the IP-multicast routing setup on your boxes. 

In this case you may want to check the routing table and make sure what
interface the IP multicast 255.0.1.1. address is bound to. When you run
Spread with no multihome config it does a universal bind to all open
interfaces (which will generally always work) but when you specify
interfaces it only binds to the ones you specify and sometimes that
doesn't match up with how the routes are setup so your traffic goes
nowhere. For example if the multicast route is through the 1.2.3.0/24
network, then in the above configuration Spread will be listening only
on the 192.168.0.0/24 network for multicasts, but they will only be sent
on the 1.2.3.0/24 network --- so nothing will work.

There has also been some past discussions of the multihoming setup 
issues and IP-Multicast issues on this list. You could try searching the 
email archives at http://lists.spread.org/pipermail/spread-users/

> 
> Any thoughts? Is what I want to do even possible with spread and wackamole?

The Spread part is definitely possible. I'm not sure if wackamole has 
any issues with the more complicated spread setup or not.

Cheers,

Jonathan


-- 
-------------------------------------------------------
Jonathan R. Stanton         jonathan at cs.jhu.edu
Dept. of Computer Science   
Johns Hopkins University    
-------------------------------------------------------