[Spread-users] send semantics

Jonathan Stanton jonathan at cnds.jhu.edu
Tue Aug 27 13:01:22 EDT 2002


The latency for a safe message to be delivered does depend on how long
it takes the token to rotate around all of the daemons. So with lots of
machines that can be high. The throughput is not impacted as
significantly. 

Fifo messages do not require as much latency as safe, but in the current
implementation they do have some dependance on the token rotation time.
So they will be faster then Safe, but will be impacted by token latency.

Some of our research papers discuss a different implementation that has
significantly improved performance for fifo messages in large
environments, and does not depend on token latency, however that is not
in the released code. 

In general, the idea of using fifo messages for many messages and
occasional safe ones to maintain some synchronization is a good idea and
can help in many applications.  

I hope this helps,

Jonathan
	
On Tue, Aug 27, 2002 at 07:34:55AM -0700, Bruce D. McLeod wrote:
> Can someone verify if I'm interpreting the
> docs and source correctly...
>   It would seem that safe messages depend
> on the token and therefore may exhibit
> scalability problems when sent from large
> amounts of nodes.
>   However...  fifo messages do not depend
> on the token and would not show the same
> scaling behavior.
> 
>   The reason I ask... I am considering using
> fifo messages for the bulk of my protocol,
> then using occasional safe messages to
> maintain coherency.  Does that seem reasonable?
> 
> Bruce McLeod
> mcleod_bruce at yahoo.com
> 
> 
> __________________________________________________
> Do You Yahoo!?
> Yahoo! Finance - Get real-time stock quotes
> http://finance.yahoo.com
> 
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users

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




More information about the Spread-users mailing list