|
Cray XD1 Research Cluster
ada.rice.edu
System News
Ada login problems week of Feb 8
posted on 2008-02-08 12:10:23
Many of you have noticed that there have been some problems with Ada's login nodes this week. The result has been that many of you have had trouble logging in, especially at night. This problem was due to the fact that there were compute jobs running on the login nodes consuming all available resources on those nodes rendering the login nodes unresponsive. This has been very disruptive to Ada's availability to all of our users.
It is not in our plans or in our philosophy to restrain resources on the login nodes. However, it is very important to be conscious of the amount of resources (i.e. CPU and memory) that your job could possibly consume. We will begin monitoring the CPU and memory usage on the login nodes in a proactive way to prevent these types of problems from occurring in the future. We will begin locking accounts of users who are either running longer than 30 minutes on the login nodes, or using too much resources (such as memory). If you have been running your code on the login nodes to test it, please begin using the Interactive queue. Details are available in the FAQ at http://rcsg.rice.edu/ada/FAQ.
Please understand that these measures are being taken to ensure the stability of Ada for the entire user community.
Do not run compute jobs on the login nodes
posted on 2008-01-16 15:26:40
please do not run compute jobs on Ada's login nodes. Running CPU intensive computations on Ada's login nodes will degrade the performance of the login nodes for all users. Our long-standing policy has been to terminate any processes on Ada's login nodes that have been running for more than 30 minutes. So please be aware that if we find any CPU intensive processes running on Ada's login nodes for more than 30 minutes, those processes are subject to immediate termination without notice.
If you are running jobs on Ada's login nodes for debugging purposes, please use the Interactive queue which was created for this purpose. Please see our FAQ at http://rcsg.rice.edu/ada/FAQ on how to use the Interactive queue.
If you have any questions on this issue, please notify the Help Desk at http://helpdesk.rice.edu.
Maximum walltime increased to 8 hours for November
posted on 2007-11-01 13:39:26
Please note that effective today (November 1, 2007) the compute queue on ada will have the maximum run time increased to 8 hours (up from 4 hours).
This will be an experiment for the month of November. We will review the impact on users and consider the best path forward in December.
Nothing special will be required of you to take advantage of this change other than requesting 8 hours instead of 4 hours in your submit script.
We are *very* interested in your feedback during this process! Please let us know ASAP if you feel this change is impacting your productivity either negatively *or* positively.
Send comments directly to Dr. Kim Andrews (kimba@rice.edu) with:
"ada policy feedback"
in the Subject Header and we will answer all email on a case-by-case basis.
Ada is back online, October 31, 10:45AM
posted on 2007-10-31 10:57:01
Ada is back online as of 10:45AM this morning, October 31. Several important tasks were accomplished during this maintenance period. Among them were some hardware repairs by Cray engineers, some important software bug fixes for Ada's internal network also provided by Cray engineers, and an upgrade of the job scheduler. The scheduler upgrade will allow us to implement some new scheduling features in the future. We will announce the new features at a later date.
|
