gomd-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gomd-devel] about the gomd+libgomd


From: Mirko Caserta
Subject: Re: [gomd-devel] about the gomd+libgomd
Date: Mon, 3 Mar 2003 22:30:09 +0100

On Mon, 3 Mar 2003 21:52:32 +0100
Ramon Pons Vivanco <address@hidden> wrote:

> > Also, as I wrote in our web page, I'd suggest using UDP for
> > "inter-nodes communication" it's going to be faster as well as
> > easier with commoncpp.
> 
> yes, we would use the socket type SOCK_DGRAM :)
> 
> The question is ... only UDP or UDP + TCP

Ok, just to see if I got the things clear... :)

1) the daemon will listen for incoming TCP connections using a
non-blocking socket (just as the openMosixInfoServer does right now);

2) if the daemon needs to grab some exotic info like /proc/uptime from a
remote node, it would UDP to the specific machine, waiting for a reply
(we probably need to set up a timeout for things like this);

3) if the daemon needs to grab some exotic info which requires more than
a bunch of raw data, then it should be using TCP to connect directly to
the remote node, probably using the standard daemon TCP interface
mentioned in point 1.

I don't think we're going to need the latest (3) since all we want to
know is already available locally from /proc, /proc/hpc and
/proc/hpc/nodes. The "exotic" things are not likely to require more than
a simple, short UDP packet (request + reply).

These are my own thoughts. Feel free to bash me for any incorrect
statement please ;)

Ciao, mc²





reply via email to

[Prev in Thread] Current Thread [Next in Thread]