Simple Distributed Computing for Scientific Research
Apple Computer, Inc.
Research and HPC Manager
Apple Computer, Inc.
Apple's Advanced Computation Group has developed a user-friendly, MPI-enabled grid management technology called Xgrid. It allows scientists to build an ad hoc cluster of Macintosh computers without complex software setup and management. It also allows two remote sites to easily enable disparate clusters or other processing tools to behave as a single coherent grid of systems.
Internet2 provides a high bandwidth infrastructure for linking disparate research sites, and Xgrid can enable these sites to easily aggregate their resources. While it does not attempt to address the complexities of the global grid, as in the case of Globus, for example, it does provide an extremely easy-to-use infrastructure for bringing trusted groups of resources together into a coherent research resource.
This poster would detail the open source and open standards technologies upon which Xgrid is based, describe Xgrid's mode of operation, and present potential collaboration opportunities within the SIG or working group frameworks of Internet2.
Intelligent Optical Networks in Research & Education
Jeff Varrant, Sr. Systems Engineer
CIENA's poster will illustrate the types of networks that are being developed in support of the Research & Education community utilizing intelligent optical networking technology. Two currently-deployed netowrks will be highlighted, the NSF TeraGrid Backbone and the I-Wire network interconnecting laboratories and university campuses in Illinois. Additionally, we'll highlight benefits associated with Intelligent Optical Networks in the Research and education community, showing the flexibility, scalability, manageability and cost effectiveness available with today's solutions.
Integrating Authentication and Authorization on Security-Demanding Environments
Fernando Frota Redigolo, M.Sc.
Tereza Cristina Melo de Brito Carvalho, Ph.D
A common scenario on many networks nowadays is the multitude of authentication and authorization (AA) systems, usually attached to the several existing applications. The use of a LDAP server is often seen as the magic solution for unifying authentication and access control information. However, as we try to integrate these applications to form a high-security environment, accommodating different authentication methods (e.g., password- or token-based) and encryption and access control regarding directory data, obstacles begin to show up. This work intends to share the experience of integrating diverse AA systems in a high security-demanding environment. Several security-related systems were tested on the course of this work, allowing an insight on how different applications use directories for AA and the difficulties in integrating them. Besides different LDAP servers, the tested applications whose LDAP support will be described include different firewalls and web proxies (for controlled web browsing), remote access servers (dial-up, VPN and RADIUS servers) and unix PAM modules, elements traditionally used as pillars of a security infrastructure.
Rich Presence Trial
PIC Working Group
Internet2 PIC Working Group members
The Presence and Integrated Communications (PIC) Working Group is engaged in a series of first light trials. These trials demonstrate SIP-based (Session Initiation Protocol) voice, video, instant messaging and voice conferencing services over an 802.11 wireless network in the context of rich presence derived from conference calendaring and HP labs location server. Trial participants download and install one of several integrated communications clients onto their laptops and PDAs allowing them to initiate communications to other participants using the receiver's email address as a single, converged electronic identity. By including rich presence services, participants are able to see not only whether other users are online or offline, but to also determine their locations and communications state.
Joint European Research Networking Development Activities
The members of the TERENA association are the National Research and Education Networks of 33 European countries. In the TERENA Technical Programme, experts from these countries work together to develop, evaluate, test, integrate and promote new networking, middleware and application technologies. In the poster session we will present some very recent results from TERENA activities. These include:
Storage Grid and Distributed Block Systems
Geoff Hayward PhD, Director of Advanced Technology
We discuss the challenges of enabling distributed storage systems and efficient, distributed data processing over distances of hundreds or even thousands of kilometers. We then review strategies for overcoming the challenges.
One strategy is to parallelize data acess and transfer across a geographically extended array of peer-to-peer storage controllers (i.e., a "storage grid"). Such a strategy can lead to dramatic improvements in data access and transfer over large distance. We describe the results of a field trial in which such a storage grid was deployed between Chicago and the Canadian cities of Vancouver and Ottawa. Different configurations of the test-bed allowed data transfers over distances ranging from 5,000 km to 20,000 km with greater than 90% utilization of the available full-duplex WAN bandwidth.
A more aggressive strategy is to deploy a 'distributed block system' through which all storage controllers in the grid export a single set of shared logical disks. Such a strategy requires that the storage controllers within the grid maintain distributed block-level coherence and all present a given LUN as a single logical device with a single, consistent data image. Typically, the strategy also requires that all servers accessing the storage grid deploy either a clustered file system, a clustered database, and/or a clustered application for accessing the shared data image.
We describe the results of two field trials in which a distributed block system was deployed in conjunction with the CXFS clustered file system over large geographic regions. Further, we describe results of emulated WAN tests in which a distributed block system was deployed in conjunction with a geographically distributed Oracle 9iRAC cluster. In presenting these results, we discuss various strategies for reducing the deleterious effects of transport latency on distributed I/O performance. In particular, we discuss: geographic caching, access-sensitive data localization, access-sensitive meta-data localization, and parallelized data access and transfer.
We argue that properly deployed distributed block systems and storage grids can provide a wide range of benefits. These include: more efficient data sharing over wide regions, reduced WAN traffic, heterogeneous storage sharing, centralized storage and archiving, and improved disaster tolerance.