This years LUG was quite interesting because it gave us a chance to get up and tell people about the Linux HPC Software stack. As a result, we''ve already got a few things that we should be thinking about how to solve (some of which we''re already thinking about, but I wanted to lay it out here): * High Availability With lustre stability and adoption increasing, a big focus has turned towards high availability. While we''ve included heartbeat, we will need to sit down and test it an really figure out the best configuration for it. * Upgrades and Package Availability Since we are currently focusing on CentOS, we have access to yum; and there was great interest in the power of being able to use a yum repository to track our packages. * Lustre Monitoring Tools With lustre, there''s a wish to bundle in the monitoring tools (e.g. collectl, lmt). * Early access to BETA There are a number of people that are very interested in gaining access to the BETA version. Perhaps we may want to toy with the idea of some kind of implemented program to allow certain people access. -- "I don''t want to be the hamster in this wheel of insanity."
Makia, Thanks for your presentation at the LUG and share below feedback. Makia Minich wrote:> This years LUG was quite interesting because it gave us a chance to get > up and tell people about the Linux HPC Software stack. As a result, > we''ve already got a few things that we should be thinking about how to > solve (some of which we''re already thinking about, but I wanted to lay > it out here): > > * High Availability > With lustre stability and adoption increasing, a big focus has turned > towards high availability. While we''ve included heartbeat, we will need > to sit down and test it an really figure out the best configuration for it.I agreed. The testing with combination of Linux RAID is also needed. As you know we have plan this.> > * Upgrades and Package Availability > Since we are currently focusing on CentOS, we have access to yum; and > there was great interest in the power of being able to use a yum > repository to track our packages. > > * Lustre Monitoring Tools > With lustre, there''s a wish to bundle in the monitoring tools (e.g. > collectl, lmt).I did a simple test that is the ''gmetric'' command in the ganglia to get the lustre status from /proc and graph them on the ganglia. We will also need further testing for the best practice and then get this configuration into the stack.> > * Early access to BETA > There are a number of people that are very interested in gaining access > to the BETA version. Perhaps we may want to toy with the idea of some > kind of implemented program to allow certain people access.
On Fri, 2008-05-02 at 22:18 +0900, Shuichi Ihara wrote:> > * Lustre Monitoring Tools > > With lustre, there''s a wish to bundle in the monitoring tools (e.g. > > collectl, lmt). > > I did a simple test that is the ''gmetric'' command in the ganglia to get the > lustre status from /proc and graph them on the ganglia. We will also need > further testing for the best practice and then get this configuration into the stack. >I have seen many references on the net to using collectl underneath gmetric... This would probably simplify the gmetric script, as well as provide a good reason for including collect. Mike
Mike,> I have seen many references on the net to using collectl underneath > gmetric... This would probably simplify the gmetric script, as well as > provide a good reason for including collect.OK, good idea and that makes sense. Thanks, -Ihara