[Spread-users] Dynamically adding a node to a group

Jonathan Stanton jonathan at cnds.jhu.edu
Wed Jul 23 23:40:29 EDT 2003


Hi,

On Wed, Jul 23, 2003 at 06:53:40PM -0700, Rob Das wrote:
> I have a quick question from a Spread newbie.  Because each Spread
> daemon requires a configuration file that lists all participating nodes
> by IP address, is there any way that a new node can be dynamically added
> to a group with out changing the configuration files for each node and
> restarting each and every daemon?

No. The configuration file must list all of the 'daemons' who could
possibly be running. However, it is easy to have extra machines
configured in the file which you are not currently using. Then when you
need to add them, you just startup the daemon with the current file and
no restarts or changes are needed. As a practical matter, some people
purposely configure extra 'machines' as ip addresses in the spread.conf
even if no hardware even exists at that IP address so they can bring it
up in the future with no restart.

There have been past discussions on this list with lots more detail
about why this is currently required. It is possible to have a more
dynamic discovery process but it will be a complex change and has some
tradeoffs.

>  
> Basically what I'm asking for is dynamic discovery of a new node where
> each existing node in the group receives a "new member" message or
> something like that.  JavaGroups does this, but I'm looking for
> something that might handle more message throughput.
>  
> Also, has anyone compared the performance of the Java binding to Spread
> vs. JavaGroups for similar underlying protocols?

I don't know of such a comparison but I would be very interested in the
results. I can provide some very good throughput numbers for Spread
using the C library. And I think we have some numbers for the
JMS4Spread JMS service that uses the Spread Java interface. 

Cheers,

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




More information about the Spread-users mailing list