Newsletter – June 2005
Optimizing your Compute Resources
NEWS/ANNOUNCEMENTS
4. DID YOU KNOW: AUTOMATED RESERVATION POLICIES
*************************************************************************
Cluster Resources, Inc. and LinuxHPC.org announced the release of Cluster Builder (www.clusterbuilder.org) — a new Web site highlighting a broad spectrum of technologies and high performance computing (HPC) information.
Cluster Builder acts as an organized repository of first- and third-party content relating to supercomputing technologies that helps alleviate the research burden of sifting through millions of search engine results. The Web site is organized into software categories such as operating systems, resource managers, compilers, etc., as well as hardware categories such as processors, switches and interconnect standards. Cluster Builder includes both open source and commercial products.
To further aid in the research, Cluster Builder directs evaluators to LinuxHPC's free Request-For-Quote service. Purchasers can use this service to complete one quote form, which is sent to multiple vendors, instead of completing a form for each individual vendor.
*************************************************************************
TORQUE* Resource Manager's user community, in cooperation with Cluster Resources, contributed several fixes in the last month to create TORQUE 1.2.0p5.
One important fix stops the problem of qstat hanging while pbs_server is waiting for an update from an offline MOM. The problem occurred when a MOM in the "down, job-exclusive" state was marked offline using the "pbsnodes -o" command.
A patch submitted by the University of Maine fixes a communication error during job cancellation. When Cluster Resources tested the patch, they found it increased the performance of qdel in certain circumstances.
On behalf of the user community we would also like to thank all those not mentioned above who helped find and fix problems and test new patches. To learn more about TORQUE visit: www.clusterresources.com/products/torque/. To join TORQUE's user community go to: www.clusterresources.com/mailing.shtml.
*************************************************************************
The Alpha version of the new Moab Grid Suite continues to receive valuable updates. The new version will include true peer-to-peer (P2P) capabilities, so local clusters can set their own policies, instead of relying on an administrator to set policies. An efficient multi-threaded I/O protocol ensures that Moab commands and P2P communication is fast and doesn't suffer from unnecessary blocking.
Another new feature in Moab Grid Suite now protects users from losing information due to the Internet's sometimes unreliable connections and bandwidth or a downed head-node. Through a new robust check-pointing feature, grid users can get up-to-date information on their grid workloads, following connectivity failure or failure on any of the peer head nodes.
Moab Grid Suite will be available later this summer. To learn more about the product, and its capabilities, visit http://www.clusterresources.com/products/moabgridsuite.shtml.
*************************************************************************
DID YOU KNOW: AUTOMATED RESERVATION POLICIES
Reservations can be created in Moab to ensure users are allowed the resources they need to complete critical tasks. However, a reservation can become problematic if the reservation creator forgets about the reservation, changes his or her plans, or reserves more resources than necessary. This leaves a set of resources that no one else can use, thereby decreasing the efficiency of cluster resources.
To combat this problem, Moab Workload Manager can be set to cancel reservations if the resources are not used within a certain time frame after the reservation begins. If a certain percentage of the resources are not used, Moab Workload Manager can also be set to reduce the amount of reserved resources. This allows others to use the resources, rather than letting them sit idle for the full reservation time alloted. The user who reserved the resources may still use them after the reservation has been canceled if they are available, but the availability is no longer guaranteed.
This feature is available through Moab Workload Manager's event policy engine. The event policy engine allows users to automate reservation policies, as well as virtually all other management policies. To learn more about reservation policies and other automated policies, visit http://www.clusterresources.com/products/mwm/docs/20.1triggers.shtml. Users can test Moab Workload Manager by downloading the Moab Cluster Suite evaluation at http://www.clusterresources.com/products/eval/.
Moab Cluster Suite® , Moab Grid Suite®, Moab Workload Manager®, Moab Cluster Manager®, and Moab Access Portal® are trademarks or registered trademarks of Cluster Resources IncorporatedTM.
* This product includes software developed by NASA Ames Research Center, Lawrence Livermore National Laboratory, and Veridian Information Solutions, Inc. Visit www.OpenPBS.org for OpenPBS software support, products, and information. TORQUE is neither endorsed by nor affiliated with Altair Grid Solutions, Inc.