From kragen@dnaco.net Mon Sep 14 22:23:59 1998 Date: Mon, 14 Sep 1998 22:23:57 -0400 (EDT) From: Kragen To: "Daniel J. Frasnelli" cc: Tracy R Reed , Shachar Tal , beowulf@cesdis1.gsfc.nasa.gov Subject: Re: Uses for a beowulf cluster? In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Keywords: X-UID: 1990 Status: O X-Status: On Mon, 14 Sep 1998, Daniel J. Frasnelli wrote: > He went on to say that multicasting node activity/status would > diminish the effectiveness of the network, and suggested a round-robin DNS > configuration. Suppose your machines are networked in a ring topology with 9600bps serial lines, and they report their status every 30 seconds. Suppose the status report is a UDP datagram containing the node's name (say, ten bytes) and its current load average (say, five bytes). The IP header is 20 bytes, the UDP header is 8 bytes, and the payload is another 15 bytes. If you're using SLIP, you have another byte for framing, for a total of 44 bytes per status report. If you have 32 machines, you'll broadcast 32 * 44 = 1408 bytes, which becomes 14080 bits with normal serial settings, every 30 seconds. That's about 1.5 seconds out of every 30 seconds taken up by status messages, or about 5%. You could include an extra 44 bytes of stuff in your status messages to bring it up to 10%. If a bigger network than you plan, on an absurdly slow physical data link, would be 5% loaded by a straightforward implementation of broadcasting node activity and status, I think you won't have a problem with network loading due to multicasting node status. Kragen -- Kragen Sitaker The sages do not believe that making no mistakes is a blessing. They believe, rather, that the great virtue of man lies in his ability to correct his mistakes and continually make a new man of himself. -- Wang Yang-Ming