[torqueusers] pbs_server and nodes file how to handle comments
garrick at usc.edu
Mon Mar 27 12:13:55 MST 2006
On Mon, Mar 27, 2006 at 01:19:43PM +0100, David Golden alleged:
> > Adding nodes almost works. It would require that the server pings all
> > of the nodes again to send out a new addr list. Since it would be
> > impractical to immediately start repinging nodes on every node create
> > operation, I was thinking about a server attribute, "ping_nodes", that
> > you would set to True to initiate a full round of node pings.
> > Additionally, newly created nodes would start in the offline state.
> > So you would 1) create as many nodes as you want, 2) 's s ping_nodes=T',
> > 3) wait for ping_nodes to become false again, 4) clear the offline state
> > on new nodes.
> Hm. Again just being too lazy to read the source, how often does the
> server publish a full node list regardless? I know for our own small
That would be a frequency of 0. New nodes start in state unknown, get
pinged, and get an addr list. The old nodes never get the new addr list.
> cluster and the frequency we find ourselves doing the operation,
> it'd be fine for added nodes to just be state = unknown (or a new state,
> like pending) for a while.
I need to also think about large clusters where a cluster-wide ping
operation can be a bit disruptive.
Garrick Staples, Linux/HPCC Administrator
University of Southern California
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torqueusers/attachments/20060327/5ab61f81/attachment.bin
More information about the torqueusers