2003 Abstracts of MCS Reports and Preprints

Reports

This document is a detailed checklist of the steps that one must go through to bring up a production computing cluster.  The list starts with planning activities and culminates in the activities necessary to operate and sustain a production computing facility. This checklist is derived from a number of experiences installing real-world, large-scale clusters.  While each installation experience was unique, we were interested in determining the common characteristics across each deployment.  We collected all of the to-do lists, presentations, notes, email messages, white board notes, and any other planning tools we could find from each of the installation activities.  We combined them into a huge, messy diagram that was probably impossible to understand without having been involved in its creation but was excellent for identifying differences and commonalities.  After organizing, checking, and distilling the information, we created the checklist presented here.  Interesting is the fact that the high-level activities on the resulting list are neither cluster nor computer specific.  Most of these activities would be followed when installing a production computer of any architecture or when installing any kind of complex facility that will eventually support users.  The purpose of this list is not to give step-by-step instructions but rather to serve as a guide and a reminder.  The items on the list are necessarily brief statements.  Detailed explanations of these would go beyond the intended scope of the list. The list is organized in outline fashion.  The major phases of construction are individual sections.  Each of the subsections is a task or subtask in that phase. The items on this list are presented in a logical sequence, in approximately the order that one would follow if one were to start with a budget and an idea.  However, every cluster is different, and every situation for using clusters is different.  Most likely, no one would ever follow the steps here in this exact order; many things can be done in a different order, simultaneously, or skipped altogether.  The list, for example, may place more emphasis on testing than many sites formally will.

Collaborative immersive virtual reality technology has been in use since the late 1980s. In the mid-1990s these systems were used to investigate multi-user wide-area collaboration scenarios. While these efforts were pioneering in many respects, they proved less suitable as work environments for everyday use. People tire easily when spending extended time in the dark spaces needed for projection virtual reality or when  being immersed in a completely synthetic world for hours at a time without access to high-resolution text displays or high-quality interactions devices. During the past decade, desktop multimedia technology capabilities have also grown considerably, to the point that all modern desktop systems can easily handle video and audio. Indeed, desktop PCs have exceeded dedicated graphics systems in rendering performance and pixel bandwidth. Moreover, with near-ubiquitous high-speed multipoint networks and protocols now available on the Internet, new models for communication are emerging. The Access Grid project builds on and extends the use of these technologies (collaborative virtual reality, desktop multimedia, point-to-point remote graphics) in ways that are better suited for users in the twenty-first century. This paper shows how we view the Access Grid as a first step toward room-based computing environments that will, we believe, challenge desktop metaphors, desktop user environments, and perhaps even desktop computer deployment in the decades ahead. Instead of simple, single-stream videoconferencing implemented with special-purpose desktop appliances, the environments we envision are entire rooms or laboratories instrumented for rich, full-time, multimodal communications between multiple groups of people connected over inherently multipoint, high-speed networks. We offer three scenarios showing how this vision is becoming a reality.

Data Grids seek to harness geographically distributed resources for large-scale data-intensive problems such as those encountered in high energy physics, bioinformatics, and other disciplines.  These problems typically involve numerous, loosely coupled jobs that both access and generate large data sets.  Effective scheduling in such environments is challenging, because of a need to address a variety of metrics and constraints (e.g., resource utilization, response time, global and local allocation policies) while dealing with multiple, potentially independent sources of jobs, and a large number of storage, compute, and network resources.  We describe a scheduling framework that addresses these problems.  Within this framework, data movement operations may be either tightly bound to job scheduling decisions or performed by a decoupled, asynchronous process on the basis of observed data access patterns and load.  We develop a family of job scheduling and data movement (replication) algorithms and use simulation studies to evaluate various combinations.  Our results suggest that while it is necessary to consider the impact of replication on the scheduling strategy, it is not always necessary to couple data movement and computation scheduling.  Instead, these two activities can be addressed separately, thus significantly simplifying the design and implementation of the overall Data Grid system.