[Spread-users] help

Jonathan Stanton jonathan at cnds.jhu.edu
Sun Sep 29 13:31:37 EDT 2002


On Fri, Sep 27, 2002 at 12:25:12AM +0800, lhs90 wrote:
> hi!
>     i want to use spread to build my wan replication database
>     but i have two problems
> 
>     First, I ran two spread daemons in the same lan and clients connect to
> different daemon join to same group can get messages, buf in wan
>     they cant. it seems like two spreads cant send message to each other.
>     In "Spread Wide Area Group Communication System", it say  "each site has
> one daemon that acts as the representative of the site"

You should just have to configure two Spread_Segments in teh same
spread.conf file and use that for daemons at both sites. Spread's
wide-area support requires that UDP traffic between the sites is not
fireawalled off or otherwise blocked. Such blocks are a common source of
difficulties. Several ports need to be open for UDP traffic between the
daemons, generally the port you configure in the spread.conf file and 1-3
more ports with the sequential port numbers. I.e if you configure Spread
to run on port 4000, ports 4000, 4001 and 4002 must also be free for UDP
traffic. 


>     i modified sample.spread.conf to my spread.conf but cant see any
> configuration about this setting.
> 
>     Second,  i know that in a lan there is a tradeoff between daemons and
> cost of recovery,  are you have any statistics about how i choose
>     a balance point or i have to experiment on this.

I'm not sure what you mean by cost of recovery -- The when a daemon
fails, all of the remaining daemons must communicate to reconstruct teh
state, so if there are more daemosn that cost is higher. The time
required should increase linearly with the number of daemons. The
absolute time required will vary with load, network performance, wan
latencies and other environmental characteristics. 

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




More information about the Spread-users mailing list