[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: "NFS Slave"??

Well, I guess I'm getting your drift now...	|:-)

Well, to put things in perspective: our number-crunching cluster is not an
isolated system, it is part of a larger set which includes central servers
and remote-boot terminals. Besides, we were running Unix here well before
Linux showed up in the scene, and pretty much took over, I might say. I am
a professor here, but have been doubling as system manager and analist for
over ten years. I currently have one system analyst and one technician,
but this small team handles three sets of machines and users similar to
this one in my Department. So, decisions and choices here are not based
solely on pure hack value, but are affected by at least: 1) history of the
system, 2) general simplicity and standardization across all systems we
manage and 3) conservation of manpower, our most scarce resource.

> Have you come across a true need for the use of NIS netgroups?

Not yet on the number crunching cluster, but plenty on our cluster of 34
remote boot terminals and the central servers. Host netgroups are used for
access control, user netgroups for sets of machines belonging to some
research group. All access control files can use netgroups: hosts.equiv,
hosts.allow, exports. This is extremely convenient to manage a large set
of machines (some 100 machines on my Department) under dynamic conditions.

> - Login information can be pushed out and synchronzied to nodes
> 	= Rsync, Scp.

Quite true! In fact, if you are using remote boot nodes, this operation
becomes a simple local copy operation on the NFS server. But since we
already had the NIS up and running when we build our number crunching
cluster, it was simpler for us to just plug them into the NIS.

> - Nodes in the cluster shouldn't be accessing outside world information
> 	= Static hosts file should be fine

This is not necessarily true even for the number crunching clusters, and
is certainly not true for terminal clusters. We use masquerading and DNS
service through it in all clusters. Some uses: mounting user homes from
machines outside; mounting a Debian mirror on a machine outside and using
apt-get for updates; doing apt-get updates via ftp.

> Excellent :) Okay, we are on the same page. knfsd is one way to go 
> for performance on linux (and even then you need to tune it).

What kind of tuning are you referring to? The transmit and receive block
sizes? Is there something else which is useful?

> Would a solution be:
> 	- Add multiple NIC's to the NFS server.
> 	- Plug it into both networks.

Yes, we tend to do this whenever possible. Is the 4 clusters we manage
(two of terminals, two of nodes) there are always 2 servers talking to
both the Internet and the private nets. But sometimes it is not possible,
like the server is in the next building or something.

        Jorge L. deLyra,  Associate Professor of Physics
            The University of Sao Paulo,  IFUSP-DFMA
       For more information: finger delyra@latt.if.usp.br

To UNSUBSCRIBE, email to debian-beowulf-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org

Reply to: