NC2 (No6) 1 Maximally Adaptive Routing Maximize adaptivity for a double-x routing based on turn model. Virtual network 0 Virtual network 1 Maximally adaptive.

Slides:



Advertisements
Similar presentations
A Novel 3D Layer-Multiplexed On-Chip Network
Advertisements

Presentation of Designing Efficient Irregular Networks for Heterogeneous Systems-on-Chip by Christian Neeb and Norbert Wehn and Workload Driven Synthesis.
Packet Switching COM1337/3501 Textbook: Computer Networks: A Systems Approach, L. Peterson, B. Davie, Morgan Kaufmann Chapter 3.
Jaringan Komputer Lanjut Packet Switching Network.
Advanced Networking Wickus Nienaber Daniel Beech.
Flattened Butterfly: A Cost-Efficient Topology for High-Radix Networks ______________________________ John Kim, William J. Dally &Dennis Abts Presented.
What is Flow Control ? Flow Control determines how a network resources, such as channel bandwidth, buffer capacity and control state are allocated to packet.
ECE 8813a (1) Non-minimal Routing Non-minimal routing  Wormhole degrades performance while VCT has less secondary effects  Fault tolerance is the main.
What's inside a router? We have yet to consider the switching function of a router - the actual transfer of datagrams from a router's incoming links to.
High Performance Router Architectures for Network- based Computing By Dr. Timothy Mark Pinkston University of South California Computer Engineering Division.
1 Lecture 12: Interconnection Networks Topics: dimension/arity, routing, deadlock, flow control.
NUMA Mult. CSE 471 Aut 011 Interconnection Networks for Multiprocessors Buses have limitations for scalability: –Physical (number of devices that can be.
1 Lecture 13: Interconnection Networks Topics: flow control, router pipelines, case studies.
1 Lecture 25: Interconnection Networks Topics: flow control, router microarchitecture Final exam:  Dec 4 th 9am – 10:40am  ~15-20% on pre-midterm  post-midterm:
1 Lecture 24: Interconnection Networks Topics: topologies, routing, deadlocks, flow control Final exam reminders:  Plan well – attempt every question.
EE 122: Router Design Kevin Lai September 25, 2002.
1 Lecture 24: Interconnection Networks Topics: communication latency, centralized and decentralized switches (Sections 8.1 – 8.5)
Issues in System-Level Direct Networks Jason D. Bakos.
1 Lecture 24: Interconnection Networks Topics: topologies, routing, deadlocks, flow control.
1 Lecture 26: Interconnection Networks Topics: flow control, router microarchitecture.
1 Lecture 25: Interconnection Networks Topics: communication latency, centralized and decentralized switches, routing, deadlocks (Appendix E) Review session,
Storage area network and System area network (SAN)
Performance and Power Efficient On-Chip Communication Using Adaptive Virtual Point-to-Point Connections M. Modarressi, H. Sarbazi-Azad, and A. Tavakkol.
Connecting LANs, Backbone Networks, and Virtual LANs
Switching, routing, and flow control in interconnection networks.
High Performance Embedded Computing © 2007 Elsevier Lecture 16: Interconnection Networks Embedded Computing Systems Mikko Lipasti, adapted from M. Schulte.
1 Lecture 23: Interconnection Networks Topics: Router microarchitecture, topologies Final exam next Tuesday: same rules as the first midterm Next semester:
1 The Turn Model for Adaptive Routing. 2 Summary Introduction to Direct Networks. Deadlocks in Wormhole Routing. System Model. Partially Adaptive Routing.
Routing algorithms Routing algorithms establish the path followed by each message or packet. Routing algorithms for wormhole routing are also valid for.
On-Chip Networks and Testing
Introduction to Network Layer. Network Layer: Motivation Can we built a global network such as Internet by extending LAN segments using bridges? –No!
Networks-on-Chips (NoCs) Basics
Dynamic Networks CS 213, LECTURE 15 L.N. Bhuyan CS258 S99.
1 Lecture 7: Interconnection Network Part I: Basic Definitions Part II: Message Passing Multicomputers.
QoS Support in High-Speed, Wormhole Routing Networks Mario Gerla, B. Kannan, Bruce Kwan, Prasasth Palanti,Simon Walton.
Deadlock CEG 4131 Computer Architecture III Miodrag Bolic.
1 Flow Identification Assume you want to guarantee some type of quality of service (minimum bandwidth, maximum end-to-end delay) to a user Before you do.
ECE669 L21: Routing April 15, 2004 ECE 669 Parallel Computer Architecture Lecture 21 Routing.
NC2 (No.4) 1 Undeliverable packets & solutions Deadlock: packets are unable to progress –Prevention, avoidance, recovery Livelock: packets cannot reach.
1 Lecture 15: Interconnection Routing Topics: deadlock, flow control.
BZUPAGES.COM Presentation On SWITCHING TECHNIQUE Presented To; Sir Taimoor Presented By; Beenish Jahangir 07_04 Uzma Noreen 07_08 Tayyaba Jahangir 07_33.
McGraw-Hill©The McGraw-Hill Companies, Inc., 2004 Connecting Devices CORPORATE INSTITUTE OF SCIENCE & TECHNOLOGY, BHOPAL Department of Electronics and.
Computer Science and Engineering Copyright by Hesham El-Rewini Advanced Computer Architecture CSE 8383 April 11, 2006 Session 23.
© Sudhakar Yalamanchili, Georgia Institute of Technology (except as indicated) Deadlock.
Interconnect Networks Basics. Generic parallel/distributed system architecture On-chip interconnects (manycore processor) Off-chip interconnects (clusters.
CS440 Computer Networks 1 Packet Switching Neil Tang 10/6/2008.
Univ. of TehranComputer Network1 Advanced topics in Computer Networks University of Tehran Dept. of EE and Computer Engineering By: Dr. Nasser Yazdani.
Networks: Routing, Deadlock, Flow Control, Switch Design, Case Studies Alvin R. Lebeck CPS 220.
CCNA3 Module 4 Brierley Module 4. CCNA3 Module 4 Brierley Topics LAN congestion and its effect on network performance Advantages of LAN segmentation in.
Virtual-Channel Flow Control William J. Dally
1 Lecture 24: Interconnection Networks Topics: communication latency, centralized and decentralized switches, routing, deadlocks (Appendix F)
1 Switching and Forwarding Sections Connecting More Than Two Hosts Multi-access link: Ethernet, wireless –Single physical link, shared by multiple.
Univ. of TehranIntroduction to Computer Network1 An Introduction to Computer Networks University of Tehran Dept. of EE and Computer Engineering By: Dr.
1 Lecture 22: Interconnection Networks Topics: Routing, deadlock, flow control, virtual channels.
Advanced Computer Networks
Lecture 23: Interconnection Networks
Deadlock.
Azeddien M. Sllame, Amani Hasan Abdelkader
Switching, routing, and flow control in interconnection networks
Bridges and Extended LANs
Storage area network and System area network (SAN)
CEG 4131 Computer Architecture III Miodrag Bolic
EE 122: Lecture 7 Ion Stoica September 18, 2001.
Lecture: Interconnection Networks
CS 6290 Many-core & Interconnect
Networks: Routing and Design
Lecture 25: Interconnection Networks
Switching, routing, and flow control in interconnection networks
Multiprocessors and Multi-computers
Presentation transcript:

NC2 (No6) 1 Maximally Adaptive Routing Maximize adaptivity for a double-x routing based on turn model. Virtual network 0 Virtual network 1 Maximally adaptive double-x (mad-x)

NC2 (No6) 2 Opt-x algorithm The Mad-x algoritm is improved by reducing the routing ristrictions on the VCs. Virtual network 0 Virtual network 1 optimal adaptive double-x (opt-x)

NC2 (No6) 3 Nonminimal routing Randomized routing (Valiant’s algorithm) is an example of nonminimal routing. It selects intermediate node randomly to balance the load. For networks using wormhole switching, nonminimal routing algorithms usually degrade performance because packets consume more network resources. So they are usually proposed for fault-tolerant routing to find alternative paths when all the minimal paths are faulty.

NC2 (No6) 4 Backtracking protocols (1/2) Backtraking protocols work on the premise that it is better to be searching for alternative paths than to be waiting for a channel to be available. From a performance point of view, backtracking protocols are suited to circuit switching. Although backtracking can be implemented with SAF switching, the performance overhead is substantial.

NC2 (No6) 5 Backtracking protocols (2/2) Most backtracking protocols use the routing header to store history information, thus increasing the size and transmission time of it. History information can be distributed throughout the nodes of networks. Each input link has a history bit vector to remember the corresponding output channel.

NC2 (No6) 6 Exhaustive profitable backtracking(EPB) It performs depth-first search of the network using only profitable (minimal) links. K-family routing protocols use a heuristic to minimize redundancy in the search for a path. In K-family, cumulative history mask of the ancestor nodes is utilized.

NC2 (No6) 7 EPB and K-family x x x x x x

NC2 (No6) 8 Exhaustive misrouting backtracking It does a depth-first search of the network using both profitable and unprofitable links x x x x x x

NC2 (No6) 9 Routing in MINs In MINs, blocking occurs when two packets must traverse the same output at a switch stage. Blocking condition for the Omega network can be tested by comparing the addresses of two input/output pairs. (equivalent conditions can be derived for other networks in a similar manner.)

NC2 (No6) 10 Blocking condition for an Omega network (1/2) To route (a n,a n-1,a n-2,…, a 1 ) to (b n,b n-1,b n-2,…, b 1 ), First stage connects (a n,a n-1, …, a 1 ) to (a n-1,a n-2,…, a 1, a n ) and a n should be equal to b n. The address of the intermediate output at each stage can be shown as the following. a n a n-1 a n-2 … a 1 b n b n-1 b n-2 … b 1 Testing for blocking between two input/ output pairs is done by the comparisons of these windows.

NC2 (No6) 11 Blocking condition for an Omega network (2/2)

NC2 (No6) 12 Switch-based networks Switch-based interconnections, such as Myrinet are used to build network of workstations or PCs. Typically, these switches support networks with irregular topologies. Such irregularity provides the wiring flexibility in LANs for scalable systems with incremental expansion.

NC2 (No6) 13 Myrinet Miricom, Inc. produced NIC and switches based on technology developed at CALTECH Univ Gb/s (Myrinet2000) Cut-through routing

NC2 (No6) 14 Myrinet switch A building block of the Myrinet switch is the 16-port crossbar switch which is mounted on a line card to connect 8 hosts with 8 ports of the backplane. Any topology is allowed.

NC2 (No6) 15 Myrinet switch enclosure Myrinet "Network in a Box" components that can interconnect up to 128 hosts with a single, full-bisection, network component, and up to tens of thousands of hosts by combining these components. Clos network for 128 hosts with all Fiber ports and monitoring capability

NC2 (No6) 16 A 32-host Clos network 8 hosts 4 links each backplane XBar16

NC2 (No6) 17 Routing of Myrinet Only the first byte of the header is examined when the packet enters a switch or interface. Once this byte is used to select an outgoing port of a switch, it is discarded (stripped) from the head of the packet.

NC2 (No6) 18 Routing in irregular topology Routing tables are required (e.g. DEC Autonet). Some mapping algorithm must be executed to fill those tables before routing can be performed. Autonet computes a breadth-first spanning tree (BFS) to fill the routing tables [Schroeder,99].

NC2 (No6) 19 An example of irregular topology switches hosts

NC2 (No6) 20 Up/down routing (1/2) Pick up one node as a root 2.Construct a spanning tree 3.Decide the up direction to the node closer to the root, or to the lower ID at the same level 4.Opposite directions against up are defined as down 5.A legal route must traverse zero or more links in the up direction followed by zero or more links in the down direction

NC2 (No6) 21 Up/down routing (2/2) Cyclic dependencies between channels are avoided because a packet may never traverse a link along the up direction after having traversed one in the down directions. It is not able to supply any minimal path between some pairs of nodes. (e.g. the shortest path from switch 3 to 2 through switch 8 is not allowed)

NC2 (No6) 22 Adaptive routing for irregular topology (1/2) Given an interconnection network and a deadlock- free routing function defined on it. Split all the physical channels in the network into two VCs, original and new. Newly injected messages can use the new VCs without any restriction as long as the original VCs can only be used in the same way as in the original routing function. Once a message reserves one of the original VCs, it can no longer reserve any of the new VCs again.

NC2 (No6) 23 Adaptive routing for irregular topology (2/2) New channels can be used to route minimal paths. Because the original channels provide nonminimal paths in most cases, newly injected messages should be leave the source switch using the new channel. (higher priority to the new channels)

NC2 (No6) 24 Resource allocation policies network resources are required to arbitrate between several packets contending for the same resource. selecting a channel or buffer among the options offered by an adaptive routing is referred as selection function. The selection function selects an output channel form the set of channels supplied by a routing function.

NC2 (No6) 25 An example of DP 89a 2 b fedc 754 ab a b Output channels are selected form a set of blue channels on each node.

NC2 (No6) 26 Selection policies Random: network state is not considered (oblivious routing) Minimum congestion: the most available VCs are selected to balance the use of them Maximum flexibility: a VC is selected in the dimension with the greatest distance Straight lines: a VC is selected in the dimension closest to the current dimension (a good distribution of traffic for 2-D meshes)

NC2 (No6) 27 Time-dependent selection function Performance may degrade if packets take long to escape form deadlock. To increase the availability of escape VCs, a selection function may delay the use of escape VCs by using a timeout. Deadlock recovery mechanisms only select escape channels after a timeout.

NC2 (No6) 28 Real-time communication Best-effort packets and guaranteed packets compete for network resources. VCs are split into two virtual networks assigning best-effort and guaranteed packets. Packets corresponding to different priority classes can be achieved by using packet switching with a scheduling mechanism (e.g. earliest deadline first).

NC2 (No6) 29 VC allocation It is required when several VCs belonging to the same physical channel are ready to transfer a flit. Random: arbitrary order Round-robin: circular order Priority scheme: some information carried in the packet header is used

NC2 (No6) 30 Throughput fairness Equal bandwidth to three flows on a shared channel Max-min fairness that the allocation to any flow cannot be increased without decreasing the allocation to a flow that has an equal or lesser allocation.

NC2 (No6) 31 Latency fairness D1A1 A2 B1 B2 C1 C2 D2 Locally fair arbitration: A1, B1, A2, C1, A3, B2, A4, D1, A5, … Age based arbitration: A1, B1, C1, C1, D1, A2, B2, C2, D2, …

NC2 (No6) 32 Non-interfering networks Allocating resources based on classes allows us to prioritize services. We may give packets of higher class strict priority in allocation of buffers and channels over packets of a lower class. To achieve isolation between two classes A and B, non-interfering networks does not allow to interrupt the usage of resources for A by B. The number of classes should be carefully chosen by the designer because of hardware complexity.

NC2 (No6) 33 Routing function unit A routing function is implemented by a single unit or several units in the input VCs. In the former case, several packets contend for the unit, requiring arbitration. Input driven: input VCs form a logical circular list and are selected in round-robin Output driven: keep a pointer to the next free output channel, and select a packet to be routed it