[Spread-users] Monitor on Win32?

[Jonathan Stanton]

On Thu, Dec 13, 2001 at 12:40:57AM -0700, Mike Stagnaro wrote:
> Thanks for the quick reply!
> 
> One more for you while I have it....
> Does performance on a ring improve with unreliable mesages versus reliable
> and greater?  Let's say that one daemon is attempting to publish a bunch of
> specific group messages at a high rate (say 1000 group messages at 1 Hz).
> Also say that 20 or 30 other daemons are listening for some mixture of these
> groups (or all of them).  Does ring performance vary depending on how these
> messages are published (reliable versus unreliable)?

With unreliable messages, the ring will not make any effort to retransmit
lost messages, while it will do retransmissions for reliable messages.
Other then that, the performance should be identical, as in both cases, one
packet is multicast on the ethernet, and once it is received by another
daemon it can be delivered. 

The distribution of groups and who is listening will not make much
difference, and all of the daemons will receive the message and then if
there are no local clients for it the daemon will discard the message. If
there are local clients, it will deliver it to them as needed. 

If one message needs to be received by several different groups, then the
message is only sent once, and anyone who is a member of any of the groups
it was sent to will receive it. So sending 1000 messages per second, where
each one goes to a different single group, or 1000 m/s where each one is
delivered to 5 groups, or 1000 m/s which all goto the same group, will cost
exactly the same.

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