Re: compute nodes, local sys calls ?
Deadpan110 <[email protected]>
| Newsgroups | gmane.linux.cluster.openmosix.general |
|---|---|
| Message-ID | <[email protected]> |
Hi, My current setup uses openMosix to loadbalance migratable processes across the cluster. All processes that can migrate will not 'fork off' and finish running on a different node. But... to do this, i also have DSH installed on every node with one designated server running cron jobs and passing the jobs to other nodes. The master server only needs the ssh keys of its worker nodes installed and so is not asked for any passwords. Once the master node runs a job on a cluster node, the processes are able to migrate across the cluster as normal and still end up finishing on the designated worker node. Links: http://sourceforge.net/projects/dsh http://howto.x-tend.be/openMosix-HOWTO/x565.html Martin On 08/02/06, Eric Stofsky <[email protected]> wrote: > > "When a process is migrated to another node, only the memory is moved. > When a command is executed, the code is loaded and begins executing on the > gateway machine. OpenMosix may then choose to migrate the process to another > node. During migration, only the executing code and data in memory is moved > to the destination node. All system calls by user's code (open, close files) > mean the process returns to the gateway to execute the system call (open, > close, get a line of data etc.). The process could migrate again once the > system call is complete." > > Is there an easy way to have all system calls (including IO) execute on > the compute nodes (i.e. not return to the gateway). Functionality, > similar to ssh'ing to each of the compute nodes and running commands, but > having openmosix controlling the "migration/remote execution". I have an > app that does lots of sequential reads intermixed with lots of computation > and the gateway is becoming a bottleneck. Note, I could easily have shared > storage mounted on each compute node. > > Thanks. > > Eric S