export disks from nodes?

Donald Becker becker at scyld.com
Thu Jan 10 15:16:59 EST 2002


On Thu, 10 Jan 2002 hanzl at noel.feld.cvut.cz wrote:

> >> - anybody figured out all the things needed for NFS server on node?
> >
> >This is the main problem, getting the daemons and the files they require
> >onto the nodes, but even that can be tricky.
> 
> I tried and got stuck on this:
> 
>  # bpsh 0 rpc.mountd --no-nfs-version 3
>  svc_tcp.c - cannot getsockname or listen: Invalid argument
>  mountd: cannot create tcp service.
> 
> ** Any explanation is more than welcome. **

I guessing that the rpc.mountd code doesn't have a default port if
/etc/services doesn't provide a port number.

In our system we omit /etc/services, along with most other /etc/*
configuration files, on the compute/slave nodes.  This typically works
great, as applications should know their correct default port.
/etc/services is (should be) only needed if you want to config a
non-standand system.

> For the rest, NFS server on scyld node seemes possible. So far I
> managed to start these (more or less related) parts:
> 
> The rpc.mountd problem might be related to resolver/names oddities,
> e.g. netstat says:
>
>  # bpsh 0 netstat          
>  Active Internet connections (w/o servers)
>  Proto Recv-Q Send-Q Local Address           Foreign Address         State      
>  tcp        0      0 .0:1592                 .-1:1936                ESTABLISHED 

A quick description of the BeoNSS name service.

BeoNSS is a cluster-specific name service that avoids the drawbacks of
existing workstation-oriented name services.  It eliminates per-node
name files, and the serialization and latency of network name lookups.
BeoNSS provides services for hosts, netgroups, password/usernames, and
ethers.

Hostnames

Cluster hostnames have the form  .<N>
This syntax was chosen because it does not conflict with valid external
(typically DNS) hostnames.

You may reference the local host as  ".-2"  or "self".
The master is known as  ".-1", with aliases of "master" and "master0".
Cluster nodes start at  .0  and extend to e.g.   .31.

A compute node has a resolvable name if has been configured, even if
hasn't been physically added.  If you configured 32 nodes and only
added 17, ".31" resolves and ".32" returns "no such host".

The currently released BeoNSS reverse-resolves the master as ".-1".
The new BeoNSS reverse-resolves the master as "master" or "master0",
with an hostname alias of ".-1".  This was done for to support multiple
masters, but improves the readability for the above case.

The name "self" never appears in reverse name resolution, and ".-2" is
mostly used in the system internals.

Netgroups

The BeoNSS system also supports netgroups, primarily for NFS servers.
The only netgroup is "cluster", with a new alias for "cluster0".  This
netgroup returns ".0" through the maximum configured slave node
e.g. ".31", even if not all 32 compute nodes have been added to the
system.

Ethers

The BeoNSS system also reports Ethernet addresses, similar to
/etc/ethers lookups.  This service is only available on the master
which maintains the cluster boot configuration.

Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list