1 / 37

Fast Broadcast in High-Speed Networks

Fast Broadcast in High-Speed Networks. Ajei Gopal, Inder Gopal and Shay Kutten. Presented By Nuthan M. Sarpangala Divya Rajesh Urmil Shah. Why Flooding protocols are unsuitable for High-Speed Networks ?. Each node - Maintains a software history of prior messages it received.

alexa
Download Presentation

Fast Broadcast in High-Speed Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Fast Broadcast in High-Speed Networks Ajei Gopal, Inder Gopal and Shay Kutten Presented By Nuthan M. Sarpangala Divya Rajesh Urmil Shah

  2. Why Flooding protocols are unsuitable for High-Speed Networks ? • Each node - • Maintains a software history of prior messages it received. • Performs a software search to determine whether or not to forward the message it received. -With increasing network speeds, software search becomes a bottleneck. Fast Broadcast in High-Speed Networks

  3. Getting around the bottleneck • Implement the search directly in very high-speed hardware. • Devise a new broadcast protocol that does not require a software search. Fast Broadcast in High-Speed Networks

  4. Objectives • Design fast distributed Broadcast protocols for a • high-speed arbitrary-topology point to point network which • Broadcasts control messages at hardware speed. • Is tolerant of failure involving loss of message. Fast Broadcast in High-Speed Networks

  5. Problem Rationale • Links are typically reliable. • Nodes have instantaneous knowledge of the state and loading conditions of links in the network. • Events like link failure, recovery & congestion occur relatively infrequently. • Dissemination time is critical in ensuring a stable network. Fast Broadcast in High-Speed Networks

  6. Assumptions • No conflicting Broadcasts. • Upper bound on the link delay is τ. • Failures are due to transient network conditions and not due to permanent removal of a node or link. • Links are FIFO and cannot generate or reorder messages. Fast Broadcast in High-Speed Networks

  7. Problem Formalization • Reliable Broadcasts: Ensures that all messages are delivered reliably and that the message traffic caused by every broadcast rapidly quiesces even under circumstances of failure. • Control Broadcasts: Weaker than reliable broadcasting, as it ensures reliable message delivery only in the absence of failure. Fast Broadcast in High-Speed Networks

  8. Reliable Broadcast • Reliable-Delivery: If a node broadcasts a message m at some time t, then all nodes deliver m by time t + L, where L is a constant called the delivery bound. • Quiescence: If m is the last message broadcast, m is broadcast at time t, then no node sends a message after time t + Q, where Q is a constant called the quiescence bound. Fast Broadcast in High-Speed Networks

  9. Control Broadcast • Delivery: If a node broadcasts a message m at some time t, and there are no failures between t and t + L, then all nodes deliver m by time t + L, where L is a constant called the delivery bound. • It also satisfies the quiescence property under all conditions. • If delivery is critical to a higher layer, it is ensured by a higher layer protocol. Fast Broadcast in High-Speed Networks

  10. Network Model Node U Node V Network Control Unit - NCUU Network Control Unit - NCUV Switching Subsystem - SSU Switching Subsystem - SS V Bi-directional Communication Links Fast Broadcast in High-Speed Networks

  11. What do SS & NCU do? SS • Message forward function is implemented in hardware. When SSu forwards m at time t, it does the following: • For all nodes v ε Neighbor(u), SSu sends m to SSv at time t. • SSusends m to NCUu at time t. • SS maintains a hardware clock and a boolean register state that can take two values – normal or exception NCU • On a request from higher level S/W to broadcast it send m to SS. • On receipt of m from SS delivers it to higher level S/W Fast Broadcast in High-Speed Networks

  12. Higher-level Software Network Control Unit (NCU) Broadcast a message Deliver a message Switching Subsystem (SS) Communication within a node and between nodes Fast Broadcast in High-Speed Networks

  13. Protocols • BASIC Protocol: -Hardware implemented • -Ensures rapid delivery and quiescence in absence of failure. • TIMER Protocol: A Control Broadcast that • - behaves like BASIC in absence of failure • - rapidly quiesces under conditions of failure. • RESPONSE Protocol: A Reliable Broadcast that guarantees • Quiescence property in case of failure & conflicting broadcasts • Ensures rapid delivery in all cases, provided network is connected. Fast Broadcast in High-Speed Networks

  14. Protocol BASIC Part executed by NCU To broadcast a message m: NCU sends m to SS On receipt of a message m´: NCU delivers m´ Part executed by SS On receipt of broadcast message m and state = normal SS forwards m state := exception SS sets timeout to expire in 2 time On expiration of the timeout state := normal Fast Broadcast in High-Speed Networks

  15. v1 v2 v3 v4 0  2 3 4 5 6 7 8 9 10 Fast Broadcast in High-Speed Networks

  16. NOTE: • Under failure free conditions protocol BASIC works because it ensures a wave of exception state nodes is created from the source outwards, preventing a broadcast message from looping back and reentering a previously visited node. • In the non terminating broadcast , SS forwards m a second time breaking the “wave” of exception state nodes, and a broadcast message is permitted to re-enter a previously visited node • If each node stays in the exception state for long enough - until every one of its neighbors has entered the exception state - the “wave” is preserved and the broadcast quiesces. • Protocol TIMER and Protocol RESPONSE both implement this intuition. Fast Broadcast in High-Speed Networks

  17. Lemma 3: Protocol BASIC does not ensure the quiescence property when processes may fail u4 u2 u1 u3 Fast Broadcast in High-Speed Networks

  18. Lemma1: Suppose there are no failures and no conflicting broadcasts, Protocol BASIC guarantees that no node forwards the same message twice Theorem 2: Suppose there are no failures and no conflicting broadcasts, Protocol BASIC guarantees both quiescence and delivery within (D+2) time Fast Broadcast in High-Speed Networks

  19. Protocol TIMER • Addresses the problem of non terminating broadcasts by forcing each node to remain in the exception state for a period of time that is larger than c where c is the number of nodes in the largest simple cycle that contains the node • c depends on a specific execution and hence is a priori unknown to each node. • Problem: node should find a appropriate approximation to c. • optimal solution : • first guess 2 and let each subsequent guess be twice as large as the previous one, for a total cost less than 2c units (in log c guesses) Fast Broadcast in High-Speed Networks

  20. Part executed by NCU To broadcast a message m: NCU sends (m,t) to SS, where t is the current clock value On receipt of a message (m,t*): T = {t´/(m,t´) was received earlier} if (T = ) NCU delivers m set a timeout to expire in the SS at time(t* +2) else t´ = max{T} set a timeout to expire in the SS at time (t* +2(t*-t´)) Fast Broadcast in High-Speed Networks

  21. Part executed by SS On receipt of broadcast message(m,-)at t and (state = normal): SS forwards (m,t) state := exception On expiration of the timeout: state := normal Fast Broadcast in High-Speed Networks

  22. Lemma 4: Suppose SS forwards m for the ith time, i>0 at time t. Then SS enters the exception state at time t and reverts to the normal state no sooner than t+2 Theorem 5: Suppose there are no failures and no conflicting broadcasts, Protocol TIMER guarantees both quiescence and delivery within(D+2) time. Fast Broadcast in High-Speed Networks

  23. Theorem 6: Protocol TIMER always quiesces Fast Broadcast in High-Speed Networks

  24. v1 v2 v3 v4 0  2 3 4 5 6 7 8 9 10 Fast Broadcast in High-Speed Networks

  25. PERFORMANCE OF TIMER Lemma 7: For any broadcast, v  V such that: 1) The maximum number of times this node is visited during the broadcast is O (log2(n)); 2) The maximum possible time interval between the first visit by the broadcast to that node and the end of the broadcast is O(nτ) Lemma 11: The maximum number of times a broadcast may visit a particular node is O(nlogn).The time for the longest broadcast is O(nτlogn) Fast Broadcast in High-Speed Networks

  26. Conflicting Broadcasts: Protocol TIMER cannot handle conflicting broadcasts. The concept of protocol Timer is extended to handle conflicting broadcasts as follows: 1. Each node maintains a set of Boolean Control Registers, ‘state-set’ instead of single register and it recognizes a set of message headers ‘fwd-set’ where |state-set| = |fwd-set| Thus by associating fwd[i] with the ith element of the fwd-set with state[i], permitting up to |state-set| conflicting broadcasts. 2. There are some context bits stored in the SS and in the header of the message ‘m’. Fast Broadcast in High-Speed Networks

  27. cont. • Now when the SS is in ‘Exception’ state,and it receives a new message, it checks for the context bits. • If the context bits were different then it would alert the software. • SS instead of rejecting this message it forwards it to the NCU • NCU checks whether this was a duplicate message or it is a new broadcast and instructs the SS to forward or rebroadcast in the case of a new message. • Thus duplicate message is never broadcast again. Fast Broadcast in High-Speed Networks

  28. FUNCTION Of NCU: To broadcast a message m: NCUu sends m to SSu On First Receipt of a message m' State’u’:= Exception NCUu delivers m' NCUu sends m' to all neighbors using a reliable data link protocol On receipt of a message m' from NCUv, v Neighbour(u); if for all w Neighbour(u), NCUu received m' from NCUw by the data link protocol State’u’:= Normal Fast Broadcast in High-Speed Networks

  29. Function of SS: On receipt of broadcast message m and state’u’= Normal: SSu forwards m State’u’ := Exception Fast Broadcast in High-Speed Networks

  30. NODE NCU SS Fast Broadcast in High-Speed Networks

  31. PROTOCOL RESPONSE • More Reliable protocol.It guarantees quiescence in the case of failure or conflicting broadcasts. • It has a stronger model than the protocol BASIC and TIMER • NCU has more CONTROL on the State Registers in this protocol and dictates when the timer expires and state should change. • When the SSu first receives a message m from its neighbor and it is in the normal state , then it FORWARDS ‘m’ to its neighbors and also to its NCUu. Fast Broadcast in High-Speed Networks

  32. Cont. • when NCUu receives the first copy of the broadcast message it too sends a copy of this to all its neighbors using a reliable data link protocol. • NCUu now waits for all its neighbors to give the copy of the message it had previously sent. • Here all the nodes sends the RECEIPT of ‘m’ instead of the complete message they received. • Only when the NCUu does not receive a RECEIPT from a particular node in a given TIME FRAME, it sends a copy of ‘m’ to that PARTICULAR NODE ONLY. • When it RECEIVES all the RECEIPT then it tells the SSu to go in a ‘normal’ State. Fast Broadcast in High-Speed Networks

  33. Cont. • Here the SSu forwards the message ‘m’ at hardware speed and the involvement of NCU comes in picture after the message has been forwarded. • Thus the primary objective of sending message at hardware speed is accomplished. Although there is an overhead required during normal flooding algorithm. • Theorem 14: If there are no failures and no conflicting broadcasts, protocol RESPONSE guarantees delivery within (D+2)τ time Fast Broadcast in High-Speed Networks

  34. CONCLUSION: • Protocol BASIC: • guarantees quiescence and delivery under failure free conditions • Protocol TIMER: • guarantees quiescence and delivery under failure free conditions • guarantees quiescence under failure conditions • handles non-terminating broadcast. • Both the above protocols can not handle conflicting broadcasts. Fast Broadcast in High-Speed Networks

  35. CONCLUSION: • Protocol RESPONSE: • It has a more stronger model as compared to BASIC and TIMER • It guarantees quiescence and delivery in failure conditions and even in conflicting broadcasts. Fast Broadcast in High-Speed Networks

  36. LIMITATIONS: * Many ASSUMPTIONS at the start make this as a weak model and is not practical to implement in the present network. * For Protocol RESPONSE, though the paper mentions about dealing with the conflicting broadcasts, it has not explained how it deals with such broadcasts. * After all such models, RESPONSE protocol looks good and stronger but then again we come to the point from where we started. - NCU has to maintain the history of all its neighboring nodes. Fast Broadcast in High-Speed Networks

  37. - It has to set an individual timer for all the nodes to which it is sending the message, so after time out if it does not receive a copy it will transmit again -NCU has to process all the details and this involves lot of time and overhead. Fast Broadcast in High-Speed Networks

More Related