[Beowulf] Naming Convention Survey

David Kewley kewley at gps.caltech.edu
Mon Jan 9 15:07:00 EST 2006

On Monday 09 January 2006 11:28, Brian D. Ropers-Huilman wrote:
> Simple question, likely a complicated answer. Cluster systems should
> have up to three distinct file systems:
> 1) a place to compile and keep "stuff"
> 2) a bit of disk to use while code is running
> 3) potential a bit of disk with special characteristics, likely
> high-performance
> Typically #1 would be called /home. Here at LSU, #2 is typically
> called /var/scratch or /var/local/scratch. Then there is #3.
> Sometimes we call this /scratch and sometimes /work.
> My question: what do other folks call these bits of disk?

Hmm.  We have a central specialized filesysteme (35TB Ibrix), plus about 
23GB free hard disk on each compute node.  /home is on the Ibrix 
filesystem.  /scratch is on the node's local hard disk.

In principle (I have yet to convince myself that this holds in practice, 
given our current conditions), a user will use /home if they want 
storage that has some of these characteristics:

* low latency (faster than local disk, it's claimed)
* high bandwidth (up to about 1GB for I/O to many large files)
* visible to all nodes
* large capacity
* intended to have more than transient use

In general, we suggest people use /home for most uses.  But node-local 
transient storage should go on /scratch.  And if a lot of nodes are 
trying to do I/O to the central filesystem all at once (many 
randomly-distributed files), at a certain attempted bandwidth or 
#files/sec, the central storage will saturate, and it may be better to 
write to local disk and collect it later.

I haven't thought out all these usage cases fully, but you get the idea.  
It bears some thought in each usage case.

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