[Beowulf] Re: Naming Convention Survey

[David Mathog]

> Simple question, likely a complicated answer. Cluster systems should
have up
> to three distinct file systems:
> 
> 1) a place to compile and keep "stuff"

/usr/common (/src, /bin, /etc, and so forth)  The compute nodes
all run programs, config files etc from in here.  No idea how common
"common" is but I chose the name as it seemed complementary to
/usr/local.  It was a pretty much a toss up between /usr/common
and /usr/cluster.  It really doesn't matter which you call it
though since the packages that go in there are mostly built
with a ./configure command and so you only need to set 
 --prefix=/usr/common (or whatever)
appropriately at build time.


> 2) a bit of disk to use while code is running

Normal users only have access to their own NFS mounted directories.
If they need to write something permanently to disk they must put
it there.

Additionally there is /scratch which is a hard disk on each node.
The primary application has a database which is split across 20
nodes and each node's piece lives in /scratch.  This is currently
read only for normal users but I do expect at some point there will be
an application which will need to be able to write in there, at which
point I will create a /scratch/tmp directory with the appropriate
permissions.

> 3) potential a bit of disk with special characteristics, likely
high-performance

/tmp on each node.

Most of the work on this system is queued up through a web interface,
runs only a finite number of programs (blast, HMMER, and related 
programs), and runs under a special account, so these details
are typically not revealed to the users.  A few advanced users
have run their own code and they used the spaces indicated above directly.

Regards,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech