 |
About DOT
DOT is a Distributed Optical Testbed that is being designed and implemented by an inter-organizational cooperative partnership to facilitate the research and development of innovative techniques that require the efficient execution of distributed applications.
The DOT research partners are creating innovative techniques required by high performance next generation applications, which are being designed to take advantage of new types of distributed information technology infrastructure, including Grid computing, advanced middleware such as Globus, and leading-edge optical networks.
Increasingly, large-scale applications are migrating from implementations based on large parallel systems at single sites to those running on highly distributed systems interlinked by high-performance optical networks. Several research communities are developing innovative high-speed optical networks and related facilities that are based on multiple 10 Gbps links (e.g., StarLight, TeraGrid, I-WIRE, and the Pacific LightRail) as well as next-generation, wavelength-switched optical networks based on advanced photonics (eg, OMNInet, an experimental testbed). These emerging infrastructures require new techniques that will allow applications to utilize efficiently distributed systems. Such systems, in contrast to parallel systems, have two unique characteristics that can be exploited: (1) heterogeneity of resources (processors and networks) and (2) dynamic changes in performance of shared resources, especially the wide area networks.
To develop robust techniques that allow applications a) to exploit these unique system characteristics and b) to take advantage of the recent work with optical networks, it is necessary to have a persistent distributed optical testbed (DOT) with a reconfigurable topology, which is a motivation for this testbed. Furthermore, it is important to explore different network topologies to explore optimizations of different forms of distributed systems.
DOT Testbed Overview
The DOT testbed consists of six sites interconnected by an optical "dark fiber" network, based on DWDM technologies. Five sites are interconnected by I-WIRE, which interlinks major research institutions and universities in Illinois, interconnects (ANL, IIT, NCSA, EVL/UIC, NU-C and UC). NU-C and NU-E are connected via OMNInet, a research network focused on studying the behavior of advanced scientific applications and how they can be integrated with and utilize advanced network management on high-performance optical networks.
- ANL: 14 data nodes
- IIT: 11 data nodes and 3 compute nodes
- NU-C: 7 data nodes and 7 compute nodes
- NU-E: 7 data nodes and 7 compute nodes
- UC: 14 data nodes
- NCSA: 8 compute nodes
- StarLight: 8 compute nodes
DOT has more data nodes than compute nodes to allow for ample storage needed by two of the research applications, ENZO and Cactus, as well as to support research focused on new data management techniques. Each data node consists of dual, Pentium IIIs with a Gbit NIC, and 180 GB disk. Each compute node is similar to the data node with the exception of having 20 GB disk instead of 180 GB disk. All of the DOT clusters are under one administrative policy, and all systems have the same software, with one scheduler used to schedule the nodes. The primary middleware used to integrate the system is the Globus Toolkit. The nodes in each cluster are connected via a router that allows for point-to-point connections. Such connections are necessary to allow the system to be scheduled as a distributed system.
|
|
