On Tue, 2004-01-27 at 14:34, Visser, Martin wrote: > All, > > As part of a project I am working on, I need to be able to potential remote control > many dozens of Linux boxes. I want to be able remotely:- > > * Create / schedule / start /stop / change running processes > * Monitor the status of those processes > * Have the processes alert me when certain milestones occur > * Have the remote boxes exchange heartbeat with the central server (so that they > know they "out of control" and need to abort any runnning jobs) > > Clearly I can do this through something like ssh/sshd and sending > scripts, or through a httpd with some cute CGI, etc. However, I prefer > a simple message framework approach that has minimal network traffic. > (I'm thinking of probably using some non-ambiguous XML message > protocol that can securely and reliably transmitted). The remote linux > boxes will be diskless and net-booted so I don't really want avoid > loading Perl or somesuch on these (even though I prefer to code in > Perl than in C).
Have you looked in to using SNMP as a message-passing framework? We use SNMP fairly extensively for remote process monitoring. It can also handle sending commands, but it's not something I've stretched it to. It took me a little while to wrap my head around the way SNMP works, and it isn't the most secure protocol (although I hear v3 addresses some of the security issues?), but it certainly covers everything else you require, and the net-snmpd package usually used on Linux boxes is pretty good. simpleweb.org is a pretty good resource for learning more about network management, and they maintain a list of SNMP-based software at http://www.simpleweb.org/software/ . Some of the free offerings look like they cover at least some of what you need in your central management server. Good luck! :-) -- Pete -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
