[Spread-users] Problem with multiple network interfaces

Vivien Quema vivien.quema at inrialpes.fr
Thu Sep 13 10:06:58 EDT 2007


Le 13 sept. 07 à 15:32, John Lane Schultz a écrit :

> Hi Vivien,
>
>
>
> First, please send the Spread_Segment blocks of your two  
> spread.conf files to the list.
>
>
>
> Second, I believe the reason why you are getting a failure when you  
> put the fast ethernet address is because Spread does a gethostname 
> () to determine the machine it is on, which probably resolves to  
> the giga address by default.  You can override this behavior by  
> running spread with the "-n <server_name>" command line parameter.   
> In this case you should just do:

> ./spread -n itaf27
>
>

It works perfectly. Many thanks.

Thank you also for all other answers. Unfortunately, I was not able  
to test them as I don't have root access on the test machines.

Just two other questions:
----------------------------------

1) I am doing benchmarks to evaluate the throughput that Spread  
achieves.

There are 3 nodes that are interconnected using a Fast ethernet  
network. Consequently, I just have one segment and each node executes  
a Spread daemon. The benchmark works as follows:

- all nodes send a burst of messages (1000 messages of 1k each).

I get around 70Mb/s, which is good on a 100Mb/s network.  
Nevertheless, what is surprising me is that I Spread achieves the  
same performance for  reliable, fifo, total, agreed and safe delivery  
semantics.

Does it make sense for you?

2) Moreover, running the same benchmark on another cluster, I get  
very bad performance (around 3Mb/s). Did you observe similar  
behaviors? I suspect these bad results to be the consequence of the  
fast switch droping multicast packets.


Thanks!
Vivien







More information about the Spread-users mailing list