I get the first part, however I also need to mention thenids for the
"internal" network. We're using fine grain routing, we have an internel
IB network and 4 seperate "lanes" of IB going out
But as I understand it, all nids need to be used with mkfs.lustre even
though no one outside of the FS server network can really see those.
On 02/03/2015 04:23 PM, Dilger, Andreas wrote:
The --mgsnode info is used by the client to fetch the config data at mount
time, and it keeps a connection open to the MGS to be notified of config
updates (including server restarts) at runtime.
The --failover or --servicenode info is sent from the servers to the MGS at
first mount to generate the config that the clients are using to mount. If a
server has multiple NIDs then the clients will try all of them if it loses the
connection during operation (i.e. if the service is failed over to another
node).
The LNet config is completely separate from the Lustre server config, because
there may be multiple Lustre filesystems sharing the same LNet network(s).
Cheers, Andreas
On Feb 3, 2015, at 12:23, Ron Croonenberg <[email protected]> wrote:
Hello,
are the options --mgsnode and --failover in mkfs.lustre (for mdt and ost) used for
"internal" purposes (so Lustre know what can be mounted where and what the
possible locations are for the MGS/MDT and OSTs ?)
Or are these values advertised, somehow, outside the file system (to the LNETs
maybe?)
thanks,
Ron
_______________________________________________
Lustre-discuss mailing list
[email protected]
http://lists.lustre.org/mailman/listinfo/lustre-discuss
_______________________________________________
Lustre-discuss mailing list
[email protected]
http://lists.lustre.org/mailman/listinfo/lustre-discuss
