[Spread-users] Problems with more than 2 hosts in a segment

[Jonathan Stanton]

Hi,

	It definitely sounds like something is wrong with your network's
ability to broadcast. 

	The 1 minute delay in Spread daemons finding each
other is because the initial ('see if anyone answers my broadcast query')
search for other daemons fails, and the daemons only find each other with
the slower ('every minute try to contact other daemons in the config
file') fallback technique. That would explain the 1 minute delay. 

	To test the broadcast with r and s you run 'r' on several
machines, and then run 's' from another machine giving the broadcast
address instead of a unicast address. You should see something like this:

[jonathan at ice5 spread-src-3.15.0]$ ./r -p 8765
Ready to receive on port 8765
-------
Report: total packets 100, total missed 1, total corrupted 0
-------


[jonathan at ice8 spread-src-3.15.0]$ ./r -p 8765
Ready to receive on port 8765
-------
Report: total packets 100, total missed 1, total corrupted 0
-------

[jonathan at ice10 spread-src-3.15.0]$ ./s -p 8765 -n 100 -a 128.220.221.255
Checking (128.220.221.255, 8765). Each burst has 100 packets, 1024 bytes
each with 10 msec delay in between, for a total of 100 packets
total time is (0,16453), with 0 problems 
[jonathan at ice10 spread-src-3.15.0]$ 

Where I ran it on ice5, ice8 and ice10 in our lab.

So here both ice5 and ice8 received the broadcast packets. If they don't
all get them then something is preventing broadcast packets.

Possible networking problems (I don't know how relevant these are because
I don't know your networking setup)

*) switch blocks broadcast
*) firewall filters
*) actual ethernet errors (check error counters on ethernet interfaces
with 'ifconfig -a')

I'm not sure exactly why 2 machines work while 3 or more don't, but it
might be because Spread sends retransmissions as unicast if only one other
machine missed them, while as broadcast if several machines missed them.
It might be that no broadcast packets work, but with only two machines the
packets are retransmitted as unicast and they work. That might explain why
it works with 2.

Hope this helps, 

Jonathan


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