by Steven Witucki, Assistant Editor
Dallas, TEXAS — “Advance Reservations” were demonstrated on the exhibit floor of SC2000 last week on two computer systems within NASA’s Information Power Grid (IPG), representing a significant step forward in computational grid technology. Three tools were used in conjunction with the IPG to obtain this achievement: Globus’ GARA, Argonne’s mpich-g2, and Veridian’s PBS Pro.
HPCwire interviewed William E. Johnston, IPG Program Director, and Bill Nitzberg, General Manager of the PBS Products department of Veridian and co-chair for the scheduling working group at the Global Grid Forum, to better understand the purpose of the advance reservation demonstration, and the technologies that enabled it.
HPCwire: What does the advanced reservation capability do for the researcher?
Nitzberg: We all take advantage of advance reservations in our ordinary lives. The classic example is air travel. Imagine if the only way to fly to Chicago was to go to the airport and wait in the standby line. Until recently, high performance computing was all scheduled in “standby” mode. This demonstration shows how researchers can reserve computer power when it is convenient for them. However, reserving computer time isn’t what’s so exciting. What’s exciting is that this is the first step towards achieving the true promise of Grids – being able to routinely use the tools of computational science in combination, regardless of geographic location or administrative ownership. Advance reservations will enable scientists to have real-time control over experiments. Imagine not just pointing the Hubble space telescope at a star, and saving the data, but reserving the compute power, network bandwidth, sky databases, and access to astronomical experts necessary to filter the data in real time, refocus the telescope, and discover water on a distant planet.
Johnston: One of the important capabilities that we expect Grids to provide is that of building systems from collections of distributed computing and data resources that will run large-scale applications “on-demand”. The process of building such virtual systems that are tailored for an application involves several steps. We must build a pool of potential resources (discovery), query that pool for a set of resources that might be co-scheduled (to support running a distributed application), and then negotiate time slots where the minimal set of resources are available simultaneously (brokering).
Very few significant resources will coincidentally be simultaneously available at the time of inquiry, so a common time in the future will most likely be the result of the negotiations of the broker. Advance reservations are the key to this, because that will provide the guarantee that when the common time slot arrives that all of the resources are still available to be formed into an aggregated system. This is true even if the time slot is the immediate future, because you will still need to ensure that all of the required resources remain simultaneously available during the time that the application will run.
With this motivation, IPG funded the development of advance reservations in the PBS workload management system.
HPCwire: What exactly are computational grids?
Nitzberg: Grids are persistent environments that enable software applications to integrate instruments, data, computational and information resources that are managed by a number of organizations in widespread locations.
HPCwire: What does grid technology provide to the scientific researcher?
Nitzberg: Grids provide scientists a uniform interface to computation resources in a similar way that a web browser provides a seamless interface to the Internet. Researchers need access to compute power, scientific instruments, storage facilities, network bandwidth, and access to scientific databases. Grid technology provides the means of accessing all these resources in a seamless manner so that the researcher does not need to be concerned with multiple protocols or different commands at individual sites, no matter the geographical distance or diverse policies that may apply.
HPCwire: Tell me about NASA’s Information Power Grid.
Johnston: NASA’s Information Power Grid (IPG) is a network of high performance computers, data storage devices, scientific instruments, and advanced user interfaces. IPG will provide NASA’s scientific and engineering communities a substantial increase in their ability to solve problems that depend on large-scale or distributed resources. IPG is part of a joint effort among leaders within government, academia, and industry, to help NASA scientists collaborate to solve important problems facing the world in the 21st century. The NAS Division of NASA Ames leads this project. For more information on IPG see http://www.ipg.nasa.gov .
HPCwire: How does IPG manage all these resources for the researcher to use?
Johnston/Nitzberg: The IPG’s “middleware” layer, with its emphasis on transparency and utility, plays a key role in the ongoing development of the Grid. This is a common layer of resources that various users of different applications and interfaces can access. This common layer then acts for the user, communicating with a heterogeneous set of resources that may be down the hall or thousands of miles away. The IPG middleware layer contains several components, including the Globus Toolkit and PBS Pro.
HPCwire: And what are Globus and GARA?
Nitzberg: The Globus Toolkit is an open source set of software services and libraries which facilitate the development of computational and data Grid applications. It implements a set of inter-operable Grid protocols which are being defined and developed within the Globus Project ( http://www.globus.org ) and Grid Forum ( http://www.gridforum.org ), for security, resource management, data management, and information services. The Globus Project is led by Argonne National Laboratory and the University of Southern California’s Information Sciences Institute, with inter-national contributions from numerous organizations including NASA, NCSA, NPACI, DOE, CERN, INFN, and many universities. The Globus Architecture for Reservations and Allocations (GARA) is a prototype of next-generation resource management protocols, services, and libraries being developed by the Globus Project.
HPCwire: What is PBS Pro and how does it fit in with this?
Nitzberg: PBS Pro, the professional edition of the Portable Batch System (PBS), is the workload management system used on both the NASA Ames and the NASA Langley IPG systems. Of interest here is the fact that PBS Pro provides the capability to handle advance reservations of compute cycles. In addition, PBS Pro interfaces with Globus to schedule work at a specified time, including work or “jobs” that span multiple systems. PBS has a history of being a leader in grid aware technologies. In 1996 and 1997 NASA used PBS as the backbone in the first production metacenter which balanced the workload across NASA Ames and NASA Langley systems using a cluster-like model of moving jobs transparently between the two centers. Check out www.pbspro.com for more information on PBS Pro.
HPCwire: Is advance resource reservation a feature that adds value to a job management system at the site level or is it’s value limited to large computational grids?
Nitzberg: It is true that much of the discussion has an eye on the need in a large scale computational grid environment – particularly situations requiring the successful coallocation of resources in order to be able to proceed with the planned computation. However, the use is not limited to that venue. There are those instances at one’s own site where the ability to reserve resources for a fixed window of time in the future in needed. For example, a researcher can better plan activities when knowing that results of a yet to be performed computations will be able to be completed and included before a publication deadline or that the resulting data will be available by a definite time. PBS Pro’s advanced reservation feature can be used with or without Globus to meet such needs.
The “advance reservation” demonstration was a joint effort of many. Dr. Jennifer Stop, Northwestern University and Argonne National Laboratory organized the demonstration, which was conducted in the Argonne booth on November 9th. The demonstration was developed by Alain Roy, University of Chicago, and Conrad Adbrecht-Buehler, Northwestern University, who supported Globus Toolkit and the Globus Architecture for Reservations and Allocations (GARA). Nick Karonis, Northern Illinois University provided assistance with MPICH-G2: the Globus enabled MPICH from Argonne National Laboratory. Also involved were representatives from NASA’s Information Power Grid and the PBS Products department of Veridian. Alain Roy performed the demonstration.