scri_info

 

scri.byu.edu

Introduction

    Scri (pronounced "scry," not "scree") is an experimental cluster of networked computers, a so-called "Beowulf."  It is our attempt to create some local parallel computing resources.  It is, in principle, available to anyone in the department that has high performance computing needs.  It should not be thought of as our own replacement for the campus supercomputers but as an additional resource within the physics department.

    Currently, scri is a network of 15 computers.  The master, or main, computer is a 1.3 GHz Athlon processor with 1.5 GB of memory.  Attached to it through a fast ethernet switch are 14 additional boxes (nodes).  Of these 14 nodes, 11 are 1 GHz Athlon CPUs with 1.5 GB of memory and 3 nodes (with 2 more to be added shortly) are dual processor machines each with two 1600+ Athlons and 2.5 GHz of shared memory.  Each node has a 20 GB hard drive yielding roughly 300 GB of total storage space.

    The main node is running RedHat 7.3 while the slave nodes are running RedHat 7.2 (with the dual nodes running 7.2 with the SMP kernel).

    We currently have the Portland Group compilers installed.

    Some comments are in order at this point.  This is not the machine to learn UNIX on.  We will expect that if you ask for an account you will be able to demonstrate a need.  The system is and will remain rather "bare-bones" in the sense that very little software beyond the operating system and standard compilers will be added.  System administration will be even more sparse so don't come asking to know how to change permissions on your files or why, after typing 'rm -rf *', things seem different on your account.  At least at first (and perhaps indefinitely) there will be few, if any, backups, so you will need to treat storage as volatile.  Indeed, you should never leave your data files on scri for any length of time if you actually want to keep them.  All work should be done on the master node and not on the slave nodes.  Indeed, code development will preferably be done primarily on your own machine with production runs taking up most of your usage of scri.

    Scri is only accessible via ssh.  Telnet, ftp, etc are all unavailable due to their inherent lack of security.

    There will be a job scheduler or batch queue provided by PBS.  We may eventually provide some basic information and perhaps some sample scripts on how to use PBS, but again, our preference is that you will have had some experience with it.

    Contact Eric Hirschmann or James Lewis for an account.