Welcome to the Scalable Computing Laboratory's informational resource for ALICE, a cost effective alternative to proprietary super computing hardware.
ALICE represents the SCL's on going investigation into methods of achieving Super Computer class performance from scientific applications in a variety of environments.
010124 User guide updates: The User Guide has been updated with documentation on the new batch system utilizing the Portable batch System and the Maui Scheduler. As of January 29th, 2001 all use of nodes 0-63 must be through the batch system!
990304 User Guide updates: The User Guide has been updated to reflect the address of the mailling list. Also, references to using DQS have been removed, and encouragement to submit/execute jobs by hand via mpirun or rsh/ssh (until a permanent queueing solution has been decided upon) was added.
980730 MPICH Buffer Size Env. Var: The following has been added to all users' default environment variables: setenv P4_SOCKBUFSIZE 65536. This increases the default buffer size for MPICH from 16k to 64k. It should result in marked improvment for message-passing performance (with the current kernel we have installed. If you would like to set this to another value, simply add the same line to ~/.cshrc.mine with the appropriate size.
980714 Improved stability: I have stepped down the kernel on all of ALICE in a couple ways. I am no longer using the 2.1 series (increased network performance), for the reasons I outlined to the mailing list awhile ago. In addition, I have removed SMP support from the "stable" kernel that is now installed. These two actions have finally stabalized ALICE to the point of useability.
I'll now be working on documentation and moving the entire machine into production mode with a queueing system and and a timesharing schedule. Once that is acomplished, I will again begin to address performance issues, first by re-enabling SMP support in newer kernels, and secondly I have a theory on how to fix the broken networking in the release series kernel (don't we all), and I'll see if I can make it a reality.
Lastly, I have been sending most of the ALICE updates/info to the list, but I'll be putting more effort into synchronizing the info here and there from now on.
980527 Downtime: ALICE will be down today while upgrades occur. The mailing list will be notified when ALICE is open again.
980515 New layout/feature: Changed the layout for the web pages, and frameified (for better or worse) them. Also added the Status link that shows the output from Brian Smith's cluster monitoring tool.
980509 Dev Node Hardware Improvements: Erratic network performance on development node64 has been solved. Also all the development nodes have been upgraded to 256MB of RAM.
980424 SCL locker on Vincent: The SCL's Project Vincent locker did not contain binaries for scp. This may have confused you if you were faithfully following the directions in the User's Guide. That has been rectified. The latest version of ssh has been installed for the AXP, DEC, and SGI platforms on PV.
980422 Status Change: The development nodes have been opened. If you filled out a policy agreement form and turned it in, you should have received mail indicating that an account has been created for you. Unlike the production nodes, the dev nodes are available for interactive use and login. This is to aid in debugging your code. The runtime environment of the development nodes is identical to the production nodes save for two differences. The development nodes have 64MB of ram (instead of 256MB), and they have only one processor (instead of two). You should utilize the dev nodes now to port your code to the ALICE environment and debug your parallel runs on 4 machines.