Nick Clarke wrote:
Hi,
-- Sorry to double post but my previous message got messed up somewhere along the way.
I stumbled across Gluster while looking for a Linux cluster system and instantly had a big smile on my face after a couple of clicks. This is a great project and not a small one.
I did have an idea for a feature, not sure if it would something for Gluster or the scheduler.
Power Save: The ability to Suspend/Sleep/Shutdown nodes (or groups of nodes) when they are not needed.
Node states:
- Suspend: Do not start up until a set time for example.
- Sleep: Faster start up
- Shutdown: Longer to start up but no power consumption
Node type examples:
- Data Centre
Achieve
Live
….
- HPC
Render Farm
Web Server
….
- Etc..
- Timed Suspend/Sleep/Shutdown
7pm – 9am for example
- Rule based power saver based on node type e.g. Data Centre, HPC node etc..
An example: A group of data centre nodes could be used for achieve and rarely used, so they could be set to Suspend after 2 hours of no use and only started when something is required. As their function is storing achieved data the user would allow for a small start up delay (common files could be cache on other nodes).
- Suspend/Sleep/Shutdown when not needed. Wake-On-Lan when needed (example: incoming large job, data access or during a time window).
This would allow me and I’m sure others to save a lot of running costs.
Nick
Nick,
I like your general thoughts on this. Cutting down the running costs is
a good goal for any cluster-based setup. Right now I would just settle
for having GlusterFS feature-complete, stable and major-bug free. But
definitely I would second adding these conservation features to the list.
Regards,
Gerry
