The computers may, for In a second embodiment of the invention each node for example, be located in a common ofEce, or alternatively joining a peer-to-peer ring-ordered dynamically reconfig­ separated from each other by great distances.

Other and further aspects and advantages of vices Digital Network (ISDN), they tend to require special, the present invention will become apparent in the following highly sophisticated cabling and circuitry.

For those skilled in the art recognize as a common interface by those skilled in the art recognize as a common interface by further example, an embodiment of an ISDN line having six further example, an embodiment of an ISDN line having six which expansion cards are connected to a computer; which expansion cards are connected to a computer; B channels and one D channels can also be used.

The present invention contemplates the addition of the variable X, then control is returned to step 46, and the a node to the network in two situations, although the entire process is repeated.

In an embodiment of the are now connected to one another in FIG. 8. invention based on ISDN, the bandwidth of the data ring is Referring now to FIG. 9, a flow chart of the preferred 50 usually 64,000 bits per second (bps) in each direction.

Levitan, and J.P. Teza The Mcube: A Symmetrical Cube Based Network with Twisted Links ______________ l l Nitin K. Singhvi and Kanad Ghose Multi-Mesh-An Efficient Topology for Parallel Processing _________________ 17 Debasish Das and Bhabani P. Sinha Accuracy vs.

A Near-Optimal Algorithm for Gossiping in ad-Dimensional Mesh Bus Interconnection Network ____ 331 Arun Jagota Session 9: Compiler Techniques Chair: Alan Sussman, University of Maryland Combining Dependence and Data Flow Anaylses to Optimize Communication ___________ 340 Ken Kennedy and Nenad Nedeljkovic Parallelizing While Loops for Multiprocessor Systems ___________________ 347 Lawrence Rauchwerger and David Padua Symbolic Range Propagation ____________________________ 357 William Blume and Rudolf Eigenmann Construction DO Loops for Non-Convex Iteration Spaces in Compiling for Parallel Machines _____ .364 Jingling Xue Compiler Techniques for Increasing CU/PE Overlap in SIMD Machines _____________ 3,69 Gene Saghi and Howard Jay Siegel Keynote Address: The New Era of MPPs: Moderately Parallel Processors Forest Baskett, Silicon Graphics, Inc.

Session 10: Parallel Architectures Chair: Jack Dennis, Massachusetts Institute of Technology The Impact of Pipelining on SIMD Architectures _____________________ 380 James D. Allen and David E. Schimmel An Assessment of COMA Multiprocessors _______________________ 388 Gyungho Lee A Workload Characterization for Coarse-Grain Multiprocessors ________________ 393 Christopher Connelly and Carla Schlatt{?r Ellis Unified vs. Split TLBs and Caches in Shared-Memory MP Systems _______________ 398 Qidong Xu and Patricia J. Teller Access Order to Avoid Inter-Vector-Conflicts in Complex Memory Systems ___________ 404 A.M. de[ Corral and J.M.

Llaberia A Unified Theory for a Traffic Analysis in Product Networks _________________ 41 l Abdelghani Bellaachia and Abdou Youssef Session 11: Scientific Computing II Chair: Vipin Kumar, University of Minnesota Geometric Mesh Partitioning: Implementation and Experiments ________________ 418 John R. Gilbert, Gary L. Miller, and Shang-Hua Teng Non-Uniform 2-D Grid Partitioning for Heterogeneous Parallel Architectures ___________ 428 Phyllis Crandall and Michael J. Quinn Toward Data Distribution Independent Parallel Matrix Multiplication ______________ 436 Hyuk J. Lee and Jose A.B.

Kaushik, C.-H Huang, J. Ramanujam, and P. Sadayappan Minimizing Communication Overhead for Matrix Inversion Algorithms on Hypercubes _______ 446 Xiadong Wang and Vwani P. Roychowdhury Folding Spatial Image Filters on the CM-5 ______________________ 451 Sandra G. Dykes and Xiaodong Zhang Session 12: Resource Management Chair: Sartaj Sahni, University of Florida Robust Parallel Resource Management in Shared Memory Multiprocessor Systems _________ 458 I-Ling Yen and Farokh B. Bastani Efficient Processor Allocation for 30 Tori _______________________ 466 Wenjian Qiao and Lionel M. Ni An Analytical Comparison of Nearest Neighbor Algorithms for Load Balancing in Parallel Computers ___________________________ 472 Chengzhong Xu, Burkhard Monien, Reinhard Luling, and Francis Lau Using Simple Page Placement Policies to Reduce the Cost of Cache Fills in Coherent Shared-Memory Systems __________________________ 480 Michael Marchetti, Leonidas Kontothanassis, Ricardo Bianchini, and Michael Scott Operating System Support for Concurrent Remote Task Creation ________________ 486 Dejan S. Milojicic, David Black, and Steve Sears Industrial Track: Invited Vendor Presentations Industrial Track Chair: John K. Antonio, Purdue University Industrial Track: Session-I Architectures and Instrumentation Chair: Richard C. Metzger, Rome Laboratory SPI: An Instrumentation Development Environment for Parallel/Distributed Systems ________ 494 Devesh Bhatt, Rakesh Jha, Todd Steeves, Rashmi Bhatt, and David Wills (Honeywell Technology Center) A Rugged Scalable Parallel System ___ __ __________ __ ___ __ _ _ .502 Alan Smeyne and John R. Nickolls (Litton Guidance and Control Systems, Inc. and MasPar Computer Corp.) The RACE Network Architecture __________________________ ~508 Bradley C. Kuszmaul (Mercury Computer Systems, Inc.) Industrial Track: Session-II Applications and Programming Chair: Sajal K. Das, University of North Texas Parallel Innovation for the Real World ______________ (paper not received at presstime) Peter Madams (nCUBE) Performance Results of Several High Performance Fortran Benchmarks _____________ 516 Douglas Miles, Larry Meadows, and Mark Young (The Portland Group, Inc.) Transformable Computers & Hardware Object Technology _________________ 518 John Schewe[, Michael Thornburg, and Steve Casselman (Virtual Computer Corporation) Session 13: Routing Chair: Assaf Schuster, Technion A Novel Deadlock-Free Routing Technique for a Class of de Bruijn Graph Based Networks ______ 524 Hyunmin Park and Dharma P. Agrawal An Efficient Scheme for Complete Exchange in 20 Tori __________________ 532 Yu-Chee Tseng, Sandeep KS.

If a portal computer is located at that search depth with a process that is fully connected to the broadcast channel, then the routine returns an indication of success.

In block 1108, the routine sets the expected number of holes (i.e., empty internal connections) for this process based on the received response.

In blocks 1804-1807, the routine loops attempting to send a connection edge search call internal message (i. e., connection_edge_search_call) to a randomly selected neighbor.

Examples of client/server middleware systems include remote procedure calls (“RPC”), database servers, and the common object request broker architecture (“CORBA”).

It would be desirable to have a reliable communications network that is suitable for the simultaneous sharing of information among a large number of the processes that are widely distributed.

The conference's success is guaranteed with a technical program of exceptionally high quality and diversity, including 192 judicially selected papers, 9 outstanding tutorials on hot and up-to-date topics, 2 panels on controversial issues, and a keynote address by Prof. Leonard Kleinrock - one of the founders of the field of computer networks.

Many leading international technology companies have located their advanced development centers in, Israel, including Lucent, Qualcomm, Sun Microsystems, Intel, IBM, Cisco, Microsoft, 3Com, and Bay Networks (now Nortel).

Thanks are also due to Gitta Abraham, for invaluable administrative assistance, Martin Peres, for printing the submitted papers, Kelly Donnelly and Ashutosh Dutta at Columbia University, who helped with the technical program meeting.

Wang, Bell Labs, Lucent Technologies Jeffrey Wieselthier, Naval Research Laboratory Geoffrey Xie, Naval Post Graduate School Opher Yaron, Congruency Ltd. Bulent Yener, Bell Labs, Lucent Technologies Ellen Zegura, Georgia Institute of Technology Hui Zhang, Carnegie Mellon University Lixia Zhang, University of California, Los Angeles Moshe Zukerman, The University of Melbourne Patrice Abry So fiene Affes Nazim Agoulmine Rajeev Agrawal Omar Ait-Hellal Cengiz Alaettinoglu Mubashir Alam Guido Albertengo D. Scott Alexander Mark Allman Yair Amir Matthew Andrews Nikos Anerousis Cosimo Anglano Farooq Anjum Furquan Ansari George Apostolopoulos Nikos Argiriou Jonathan Amold Kostya Avrachenkov Yossi Azar Dennis Baker Mario Baldi Anindo Banerjea Subrata Banerjee Sujata Banejee Jozsef Barta Anindya Basu Andy Bavier Jochen Behrens Bhargav Bellur Shai Benjamin Arthur Berger Azer Bestavros Andre-Luc Beylot Vaduvur Bharghavan Supratik Bhattacharyya Giuseppe Bianchi Ezio Biglieri Chatschik Bisdikian Erez Bitton Marcel0 Blatt Noam Bloch Chris Blondia Jean Bolot Ethendranath Bommaiah Thomas Bonald Flaminio Borgonovo Sem Borst Khaled Boussetta Christian van den Branden Reviewers Jack Brassil Anat Bremler-Barr Lee Breslau Josh Broch Zvika Bronstein Patrick Brown Marcus Brunner Jose Brustoloni Tian Bu Milind Buddhikot Ken Budka Santithom Bunchua Dennis Bushmitch James Cai Franco Callegati Ken Calvert Fraser Cameron Tracy Camp Antonio Capone Jeffrey Capone German0 Caronni Christopher Carothers Claudio Casetti Coskun Cetinkaya Kaushik Chakrabarty Sammy Chan Shueng-Han Gary Chan Cheng-Shang Chang Jae-Hwan Chang Yuan-Chi Chang H. Jonathan Chao Hao Che Biao Chen Shiwen Chen Tom Chen Yao-Min Chen Shun-Yan Cheung Carla Fabiana Chiasserini Wonhong Cho Jinwoo Choe Edwin K. P. Chong Tim Chow Yanghua Chu Israel Cidon Jorge Cobb Denis Collange M. Scott Corson Scott Corson Irene Cozzani Mark Crovella Luiz A. DaSilva Ricardo Dahab Apostolos Dailianas Sajal Das Samir Das Debasish Datta Michel Diaz John Dilley Gopal Dommety Jeff Donahoo Lakshminath Dondeti Constantinos Dovrolis Peter Druschel Andrzej Duda Nick Duffield Ashutosh Dutta Zbigniew Dziong Eylem Ekici Georgios Ellinas Anwar Elwalid Anthony Ephremides Ozgur Ercetin Funda E r g y Elza Erkip Cem Ersoy Deborah Estrin Ted Faber Sonia Fahmy Aiguo Fei Wu-chi Feng Pablo Molinero Femandez Enrica Filippi Greg Finn Danilo Florissi Sally Floyd Chuan-Heng Foh Nelson Fonseca Andrea Francini Avraham Freedman Oleg Frenkel Timur Friedman Victor Frost Chane Fullmer Andrea Fumagalli Rossano Gaeta Mike Gallagher Aura Ganz Jian-Bo Gao Jun Gao Lixin Gao Fatima Garawi Juan Garay Rodrigo Garces Eric Gauthier Constant Gbaguidi Evaggelos Geraniotis Panos Gevros Dipak Ghosal Atanu Ghosh Amit Gil Stefan0 Giordano Andre Girard Maxim Gitlits Nada Golmie Leana Golubchik Javier Gomez Agusting Gonzalez Gilad Goren Sergey Gorinsky Shahar Gorodesky A. Gosh Ramesh Govindan Rohit Goyal Vivek Goyal Marcel Graf Bob Gray Albert Greenberg Timothy Griffin Sudipto Guha Ajay Gummalla Oktay Gunluk Katherine Guo Yang Guo Minaxi Gupta Pankaj Gupta Omer Gurewitz Ani1 Gurijala Robert Haas Zygmunt Haas Mounir Hamdi Mark Handley Stephen Hanly Ruibing Hao David Harrison Oded Hauser Llnhai He Gerard Hebuteme John Heidemann Abdelsalam Hela1 Debra Hensgen Gisli Hjalmtysson Joseph Ho Lawrence Ho

David K. Y. Yau Zhong Ye Aylin Yener Pinar Yilmaz Lany Younkins Alaa Youssef Xiaoping Yun Daniel Zappala Bill Zaumen William Zaumen Danlu Zhang Kevin Zhang Lisa Zhang Zhensheng Zhang Zhi-li Zhang HuiMin Zhang Li Wei Zhao Weibin Zhao Weihua Zhuang Thomas Ziegler Michele Zorzi Li Zou Srinivasan Rao Muthukumar Ratty Danny Raz Gianluca Reali Amy Reibman Martin Reisslein Reza Rejaie Qiang Ren Tianmin Ren Shai Revzen Luigi Rizzo Philippe Robert Vincent Roca Charles Rohrs Timothy Roscoe Christopher Rose Marcel Rosu Jean-Louis Rougier George Rouskas Rajarshi Roy Sumit Roy Roberto Sabella Bahareh Sadeghi Debashis Saha Cenk Sahinalp Sambit Sahu Jakka Sairamesh Kave Salamatian Theodoros Salonidis Medy Sanadidi Ricardo Sanchez Iraj Saniee Brunilde Sans0 Saswati Sarkar Baruch Schieber Christoph Schuba Paolo Scotton Panagiotis Sebos Adrian Segall Nemo Semret Subhabrata Sen Sudipta Sengupta Ahmed Serhrouchni Bruno Sericola Dimitrios Serpanos Srinivasan Seshan Harish Sethu Moshe Shahar Anees Shaikh Shervin Shambayati Vishal Sharma Oran Sharon Jacob Sharony Scott Shenker Sherlia Shi Chih-Heng Shih Nahum Shimkin Rajeev Shorey Sandeep Sibal Lionel Silman Suresh Singh Rakesh Sinha Vasilios Siris Dorgham Sisalem Krishna Sivalingam Vijay Sivaraman Paul Skelly Khosrow Sohraby Arun K Somani Edmundo de Souza e Silva Oliver Spatscheck John Spinelli Alexander Sprintson Kunwadee Sripanidkulch Kotikalapudi Sriram Mani Srivastava Rolf Stadler David Starobinski Ioannis Stavrakakis Peter Steenkiste Donpaul Stephens James Sterbenz Ion Stoica Sasha Stolyar Chi-Jiun Su Suresh Subramaniam Devika Subramanian Madhu Sudan Wei Sun Behrnard Suter Boleslaw Szymanski Vahid Tabatabae Marco Tacca Avner Taieb Hideaki Takagi Yoshitaka Takahashi Nisha Talagala Rajesh Talpade Fabrizio Talucci Ken Tang Asser Tantawi Renu Tewari Chen-Khong Tham Patrick Thiran Srinivasa Thirumalasetty George Thomas Mikkel Thorup Xusheng Tian Terence D. Todd Samir Tohme Panos Trimintzios John Trotter Christos Tryfonas Vassilios Tsaoussidis Thierry Turletti Rafi Tzadik Guillaume Urvoy Elif Uysal Andras Valko Bobby Vandalore George Varghese Anujan Vama Vasos Vassiliou K. R. Venugopal Veronique Veque Vinicio Vercellone Dinesh Verma Sanjeev Verma Olivier Verscheure Joan Viaplana Lorenzo Vicisano Brett Vickers Anastasios Viglas Curtis Villamizar Patrick Vincent Yannis Viniotis Jorma Virtamo Kirankumar Visa Pramod Viswanath Emanuele Viterbo Marcel Waldvogel Fugui Wang Huayan Wang Shieyuan Wang Weiguo Wang Wenye Wang Xin Wang Zheng Wang Andy Ward Kimberly Wasserman Vijitha Weerackody Philip Whiting Walter Willinger Stephen Wilson Damon Wischik Chung Kei Wong Vincent Wong Thomas Woo Lloyd Wood Avishai Wool Rebecca Wright Tao Wu Ye Xia George Xylomenos Anlu Yan Shih-Tsang Yang Xiaowei Yang Roy Yates Table of Contents Volume 1-Tuesday Ad Hoc Networks Performance Comparison of Two On-Demand Routing Protocols for Ad Hoc Networks ............................................ S .

Du, New Brighton, both of Minn. Minnesota, St. Paul, Minn. "Dense Trivalent Graphs for Processor Interconnec "A Survey of Interconnection Networks", Tse-yun [73] Assignee: tion”, Leland and Solomon; IEEE Transactions on Regents of the University of [21] Appl.

Hence, this tion between two microprocessors, dedicated pins on 40 shorthand notation is a convenient method of represent- tion between two microprocessors, dedicated pins on shorthand notation is a convenient method of represent the LSI package are used to form the communication ing larger and more complex hardware multiprocessor the LS1 package are used to form the communication ing larger and more complex hardware multiprocessor interconnect.

These types of invention, the comers of the modified hypercube are prior art direct interconnection techniques are very invention, the corners of the modi?ed hypercube are prior art direct interconnection techniques are very substituted with topologies having cycle paths in order substituted with topologies having cycle paths in order expensive for loosely-coupled or distributed multipro­ expensive for loosely-coupled or distributed multipro to reduce the degree of the overall structure.

With this view, much current com neighbors to keep the degree small has been proposed in puter architecture research has focused on the use of neighbors to keep the degree small has been proposed in puter architecture research has focused on the use of identical processors in homogeneous configurations 60 the prior art in the form of several different topologies the prior art in the form of several different topologies identical processors in homogeneous con?gurations for the family of trivalent graphs.

This embodi CM-1 is constructed from a custom designed VLSI chip CM-] is constructed from a custom designed VLSI chip ment is described in sufficient detail to enable those ment is described in sufficient detail to enable those that contains 16 processor cells and one router unit of that contains 16 processor cells and one router unit of skilled in the art to practice the invention, and it is to be skilled in the art to practice the invention, and it is to be .

Saikat Guha, Alexey Reznichenko, Kevin Tang, Hamed Haddadi and Paul Francis, “Serving Ads from localhost for Performance, Privacy, and Profit,” In Proceedings of the 8th Workshop on Hot Topics in Networks HotNets ’09, 2009 0002

Changxi Zheng, Lusheng Ji, Dan Pei, Jia Wang, Paul Francis, "A Light-Weight Distributed Scheme for Detecting IP Prefix Hijacks in Realtime," ACM SIGCOMM 2007, August, 2007, Kyoto Japan 21.

Manpreet Singh, Prashant Pradhan, Paul Francis, “MPAT: Aggregate TCP Congestion Management as a Building Block for Internet QoS” 12th IEEE International Conference on Network Protocols (ICNP 2004), October 2004.

Paul Francis, Sugih Jamin, Vern Paxson, Lixia Zhang, Daniel Gryniewicz, Yixin Jin, "An Architecture for a Global Internet Host Distance Estimation Service," Proceedings INFOCOM '99, New York City, March 1999.

Saikat Guha, Bin Cheng, Alexey Reznichenko, Hamed Haddadi, Paul Francis, “Serving Ads from localhost for Performance, Privacy, and Profit,” 8th ACM Workshop on Hot Topics in Networks, Hotnets 2009, New York, Oct. 2009 52.

The choice of threshold val ues for capacity increase and decrease affects response time as well as sta bi Ii ty, and should possibly be cha nged dynamically on a per-YPC basis in order to adapt to usage patterns.

l n orde r to mainta in the cell level performance of existing VCCs, logical bandwidth modifi cat ion is don e fir st so that the occ up ie d capacity ca n be decreased as calls are te rminated.

Triggering Mechanism A ltho ugh mu ch wor k has been devoted to d esignin g a VPC overlay network with optima l capacity assignme nts, li ttl e is said about what wo uld actu ally tri gger th e reco nfi gurat ion p rocess.

IEEE Network • September/October 1996 Finally, Logothetis and Sh ioda (47] propose an activity where global capacity realloca ti o ns are made (wit hout accompany ing topology cha nges) whe n the observed call blocking probability exceeds some threshold.

The second optimization problem then finds a VPC overlay netwo rk to accom modate these routes that minimizes call setup, switching, and transmission cos ts by u sing sh orter pat hs (i.e., with fewe r V P C hops) between source -destination pairs with high traffic demands.

Dealing with dynamic topologies Yallcast Tree Protocol (YTP) (framing, forwarding, sequencing) yTCP yRTP yRMTP Etc... Getting through NAT boxes Yallcast ID Protocol (YIDP)

Member Discovery • Learn of other members to build topologies – Contact Rendezvous (initial) – Parent-side tree anycast or mesh anycast discovery messages (background activity) – Navigate tree (as-needed foreground activity) – If root, broadcast “I am root” message, inform rendezvous 0009

Root Path • Analogous to BGP AS-path • Transmitted parent to child – Each member appends itself to received root path • Discovers loops after tree change 0011

How Loops Form “Simultaneous” changes by multiple members (e.g. P and L)

Coordinated Loop Avoidance • Send “Intent to Join” along tree from joining member to new parent • Members along the path record intended change • Members along the path check for and block incompatible changes 0020