Special Topics in Computer Engineering Instructor: Dr. Walid Abu-Sufah

1. Special Topics in Computer EngineeringInstructor: Dr. Walid Abu-Sufah Name: Nasser Anssari University No: 0037253

2. Network Design:Wireless Networks

3. Vision Of Research • A generation of ad hoc networks whose performance plane is on par with wired networks in terms of latency, capacity and robustness. Such a capability would enable an emerging breed of high-bandwidth real-time multimedia applications to run over large ad hoc networks.

4. Outline • RAIN: A Reliable Wireless Network Architecture • Challenges: A Radically New Architecture for Next Generation Mobile Ad Hoc Networks • A Cautionary Perspective on Cross-layer Design

5. RAIN: A Reliable Wireless Network Architecture Chaegwon Limyz, Haiyun Luoy, and Chong-Ho ChoizUniversity of Illinois at Urbana-Champaign, USASeoul National University, Korea 2006

6. Routing in MANETs • Routing and transport protocols for wired, first-/last-hop wireless Internet, are often extremely unstable and unpredictable in a multihop wireless network as the network grows, especially when the offered load is high. • Excessive packet drops • Unfair channel bandwidth sharing and starvation • Extremely volatile path properties

7. Why Do Such Problems Occur? • Wireless transmissions are broadcast in nature: • Contend with each other even between packet transmissions of the same flow. • Interfere with each other in a range that is usually longer than the transmission/receiving range and unknown a priori. • Hard for wireless MAC to coordinate interfering nodes with which it cannot directly communicate. • The contention becomes the primary cause to packet losses in a multihop wireless network as compared with the wired Internet where packet losses are mostly caused by network congestion and router buffer overflow.

8. A Solution Approach • Wired Internet routers are equipped with more and more buffer to mitigate buffer overflow problem caused by traffic bursts. • On the contrary to this common wisdom, in multihop wireless networks the length of the transit traffic queue should be controlled to the minimum for effective contention and congestion control.

9. How Would Buffer Management Limit The Contention Level? • Since all wireless nodes with non-empty packet queues will contend for the channel, the contention level is a function of the number of nodes with backlogged queues. • Reducing the number of backlogged queues in the network would keep the contention level low. • To further prevent multiple flows from overwhelming the short queues at intermediate routers, enforce congestion control through in-network implicit back-pressure to extend the scope of the contention control at an individual wireless router’s MAC layer all the way to the packet sources.

10. RAIN Architecture

11. RAIN Architecture (Cont.) • Preserve the layering architecture of the Internet, but redefine the service models for the datalink layer, the network layer, and the transport layer. • The most notable difference is that in RAIN the contention control is added and the congestion control is pushed down the stack to the datalink layer.

12. RAIN Architecture (Cont.) • The network layer populates the routing table and provides the default route lookup. The network layer transit traffic buffering is united with the datalink layer. The network layer still maintains a buffer, but only for packets generated by or destined for the node. • The transport layer may detect and recover packets that are lost due to router or routing failures, in case perfect reliability is desired.

13. RAIN Datalink • RAIN’s contention and congestion control can be implemented using simple datalink layer mechanisms. • SAFE: small buffer and adaptive freeze: based on the IEEE 802.11 standard. • Assuming: • A wireless transceiver performs physical and virtual carrier sense before transmitting a frame. • Per-frame acknowledgement is enforced.

14. Contention And Congestion Control • SAFE datalink maintains a single buffer (FIFO), called transit traffic buffer, for all outgoing frames including those generated by the node itself. • If the Freeze Region is empty, the node behaves as a normal 802.11 device. Once frames start to be buffered in the Freeze Region, the contention and congestion controller returns a contention and congestion control signal, piggybacked to the per-frame acknowledgement, to the upstream sender.

15. Contention And Congestion Control (Cont.) • The freeze region is used to accommodate the overflow of the small free region due to signaling errors. • In the rare cases the buffer is indeed full, due to either signaling error or small buffers at low-end transceivers, SAFE contention and congestion controller sends back a negative acknowledgement to the upstream wireless router. The upstream router that receives such a negative acknowledgement will then retransmit the frame after appropriate freezing.

16. Contention And Congestion Control (Cont.) • In summary, through SAFE datalink layer contention and congestion control a RAIN network can effectively minimize the contention and completely eliminate packet losses due to buffer overflow. On the contrary to the common belief, it is possible to build a multihop wireless network that is even more reliable than a wired network, using off-the-shelf CSMA/CA wireless transceivers.

17. SAFE Frame Format • If transit traffic buffer status is set, i.e., the Free Region is full, the requested freeze duration is set to (Tpkt * N / 100µs) where Tpkt and N are the average transmission time for a data frame and the number of data frames in the transit traffic buffer respectively. • If the transmit traffic buffer status field is not set, then the requested freezing duration field must be zero. Any non-zero freezing duration together with the transit traffic buffer status unset represents a negative acknowledgement

18. Medium Access Control • SAFE maintains a Table of Neighbor Status (ToNS) withone neighbor per entry. • The table entry is updated whenever a wireless router receives or overhears the two SAFE control fields • The way SAFE enforces the requested contention and congestion control is similar to the way virtual carrier sense works in the IEEE 802.11 standard. Therefore building SAFE into the existing CSMA/CA wireless transceivers will be straightforward.

19. RAIN Transport • A packet can be lost in a multihop wireless network due to the following five reasons: • Wireless channel errors • Excessive contention for the shared wireless channel • Congestion at intermediate wireless routers • Wireless router failures • Routing failures due to high node mobility. • Packet losses due to channel errors have been addressed in standard CSMA/CA wireless MAC through channel rate adaptation, per-frame acknowledgement and retransmission. • With SAFE datalink layer implemented in a RAIN network, packet losses due to contention or congestion are highly unlikely.

20. RAIN Transport (Cont.) • Therefore, for applications that demand 100% reliability, a reliable transport layer only needs to deal with packet losses due to failed wireless routers or routing. • A full-fledged TCP has are a number of TCP functions unnecessary in a highly reliable RAIN network. • When node mobility is modest, e.g., in the wireless mesh network, router and routing failures are both rare. The transport protocol can exploit this fact, and acknowledges data segments less frequently to further improve the end-to-end throughput. • ReTP: reliable transport protocol designed specifically for a RAIN network.

21. ReTP vs. TCP

22. Performance Evaluation • Grid Topology

23. Performance Evaluation (Cont.)

24. Outline • RAIN: A Reliable Wireless Network Architecture • Challenges: A Radically New Architecture for Next Generation Mobile Ad Hoc Networks • A Cautionary Perspective on Cross-layer Design

25. Challenges: A Radically New Architecture For Next Generation Mobile Ad Hoc Networks Ram Ramanathan Internetwork Research Department, BBN Technologies, Cambridge, MA, USA 2005

26. Why Does The Performance Of MANETs Lag Behind That Of Wired Networks? • A key reason for this deficiency is that all current-generation MANET systems share some basic architectural features that result in severe under-utilization of the performance potential. • The operations are hop-centric in that processes are terminated and re-initiated at every hop. This has a drastic impact on performance, not only on path latency but also on effective capacity.

27. Hop-Centric Operation • The packet has to be processed at three layers for header stripping etc., re-queued at the network and MAC layers, processed again for header insertion, and re-contend for channel access, with the obligatory backoff and retransmission. And this happens all over again at the next node, and at every single node along the way to the destination.

28. MANET Under-Performance (Cont.) • The physical layer used in a MANET node is ill-suited for multi-hop or relay-based communications. The current MANET physical layer is optimized for two primitives – to receive and to transmit, whereas in MANETs the most common operation, and one that is the essence of wireless multi-hop networking, is relaying. • Current architectures fail to take advantage of the broadcast nature of MANETs. For instance, when a node transmits a packet, it delegates a single neighboring node to retransmit the packet. All other neighboring nodes that receive the packet are made to discard it.

29. A Notional Performance Requirement • A network with 10,000 or more mobile ad hoc nodes, diameter (and hence path lengths) of 50-100 hops, transport capacity of 1 Gbps, end-to-end latency less than 10 ms, and wireline-like robustness. • Current MANET architecture has scope for incremental advances, not the order-of-magnitude gains needed to meet such challenging requirements. A new generation of MANETs with a radically new architecture is required.

30. A Vision For The Next Generation MANET Architecture • A physical layer that is optimized for multi-hop wireless networking • Access to the medium for the entire path (Path-Centric) • Cooperative transport of packets • Executing the above vision would require a re-design of the transceiver, in particular making specialized chipsets for MANET nodes distinct from chipsets for single-hop communications.

31. Relay-Oriented Physical Layer • A key part of MANET control is determining which set of nodes relay the packet from the source to the destination (routing), and transporting along this chosen path (forwarding). Currently, both of these are done two layers above the physical layer. • Both functions (jointly called relaying) are moved to the physical layer, a third primitive on par with transmitting and receiving. • Both unicast and global broadcast packets are switched at the physical layer itself, without the involvement of any of the higher layers.

32. Path-Centric Operation • End-to-end transport does not involve the termination and subsequent re-initiation of the access protocol procedures at every intermediate hop. • Extend the conventional contention-based floor acquisition (using RTS-CTS) to be path-oriented. • Medium access control is path-oriented in that access to the channel at the source is for multiple hops enroute to the destination and the packet does not have to recontend at each intermediate hop.

33. Cooperative Transport • Opportunistically harness unused resources to increase the capacity of a path. • The emerging concept of cooperative diversity is a powerful method for doing this at the physical layer. In cooperative diversity, nodes simultaneously retransmit the same packet to be diversity-combined at receivers. • A “band” of nodes between the source and the destination cooperatively guide an energy conduit by re-energizing and combining to create a high-capacity, highly robust pipe.

34. Putting Things Together

35. Putting Things Together (Cont.) • The three key elements of the architecture – relay-oriented physical layer, path access control, and cooperative transport are inter-related and synergistically combine to yield benefits in excess of their individual value. • Path access control is much faster because its control packets are relayed at the physical layer. • Physical layer switching is possible only because the channel is reserved up front by path access control. • Cooperative diversity is enabled by having relaying at the physical layer. • Routing control packets use cooperative diversity for fault tolerance.

36. In What Way Will This New Vision Improve Performance? • Latency The dramatically reduced processing and elimination of re-contending at every hop will significantly reduce latency. • Capacity Capacity will be significantly increased by virtue of cooperative transport and switching at the physical layer by eliminating the delay bottleneck, allowing path capacity to better approximate link capacity. • Path Reliability Path diversity along with the use of cooperative transport significantly increases path robustness. • Energy Energy consumption will be reduced due to the reduced processing and contention, and the use of cooperative diversity.

37. Design Approach • Relay-oriented Physical Layer • A relay-oriented physical layer should have the ability to transmit and receive packets simultaneously (full-duplex operation). • Simultaneous transmission and reception can be achieved using multiple frequencies or orthogonal codes, with the transmitter and receiver tunable to different, non-interfering frequencies or codes. • Relaying consists of two problems: routing, that is, deciding which node(s) do the relaying, and forwarding, that is, the actions each node should perform to actually make the packet go along the chosen path.

38. Design Approach (Cont.) – Relay PL • Forwarding:

39. 5 3 5 2 2 1 3 1 2 1 A D B E C F Design Approach (Cont.) – Relay PL • How to decide the sequence of nodes (route) that will relay the packet? Use a proactive link-state routing approach. • Link-State Routing Algorithm (Dijkstra’s Algorithm): • Computes least cost paths from one node (‘source”) to all other nodes • Accomplished via “link state broadcast” • Gives routing table for each node • All nodes have the same information

40. Design Approach (Cont.) – Relay PL • The proposed routing mechanism adapts link-state routing to run at the physical layer, that is, the routing updates and neighbor discovery probes, if any, do not use the MAC layer. • Link State Updates (LSUs) are generated in a conventional manner when a link goes up or down. • The flooding of a generated LSU involves a network preamble followed by the actual LSU. The network preamble is a multihop preamble that gets the nodes to ignore data transmission or reception and tune into the ensuing LSU. • The preamble and the LSU are relayed (re-broadcast) at the physical layer – the RX coherently combines multiple updates using diversity of transmission from various nodes.

41. Design Approach (Cont.) – Path Access Control • Extension of the 802.11 DCF MAC protocol to a path-centric regime. • Source s1 wants to transfer a burst of data (one or more packets) to a destination d1: The PAC in s1 sends a Segment Access Request (SAR) which gets relayed at the physical layer until it encounters an ongoing transfer. • For this s1-d1 transfer, assume that the SAR makes it all the way to d1, without encountering an ongoing transfer and without exiting the physical layer.

42. Design Approach (Cont.) – Path Access Control • The PAC in d1 replies with a Segment Access Clear (SAC), which again is relayed through the physical layer and reaches s1. • The SAR can be thought of as a “multi-hop RTS” and the SAC as a “multi-hop CTS”, switched at the physical layer. • Upon receiving the SAC, a data burst (one or more packets) is relayed. • An acknowledgement may be sent, or we may simply let the transport layer take care of reliable delivery.

43. Design Approach (Cont.) – Path Access Control • The nodes that overhear the SAR or SAC create state to protect the transfer. Such nodes are called the sentinels for the transfer (path-centric equivalent of the Network Allocation Vector (NAV) in the 802.11 DCF ) • Consider source s2 wishing to send to destination d2, while the s1-d1 transfer is in progress: • The SAR from s1 hits a sentinel (W) and is prevented from going further. • The SAR is delivered to the PAC at this intermediate node W, terminating a segment. • A return SAC clears the way for data burst transfer over this segment. • Node W can use another route to set up the next segment going “around” the sentinels, or it can wait until the s1-d1 transfer completes, and use a more direct path.

44. Design Approach (Cont.) – Path Access Control • Packets may be transmitted without a preceding SAR/SAC, just like the DATA-ACK mode in 802.11 DCF does not use RTS/CTS. In such a case, each packet behaves as the SAR does. • How to set up of the frequencies of each node’s RX and TX to enable full-duplex operation? • Select a TX frequency along with the transit decision and let the RX “auto-tune”, i.e., sense on all frequencies and tune to the one on which energy is the most. • Use SAR/SAC and transfer them on an apriori assigned frequency in half-duplex mode, and set the RX and TX frequencies appropriately as they go for the ensuing burst.

45. Design Approach (Cont.) – Cooperative Transport • Routing along a single path, as is mostly done currently, fails to exploit the richness of the topology and the additional resources (e.g. power) of other nodes that could potentially help in the packet transport. • MANETs offer the “wireless broadcast advantage” – a single transmission reaches many nodes, and “wireless cooperative advantage” – multiple nodes can cooperatively transmit the same packet that could be exploited on an end-to-end basis. • Cooperative diversity is the near-simultaneous transmission of the same information by multiple nodes that is coherently combined at the receiver.

46. Design Approach (Cont.) – Cooperative Transport • Node S broadcasts a packet that is rebroadcast near-simultaneously to make forward progress toward D. Nodes that do not contribute to forward progress discard the packet. Nodes that do contribute to forward progress receive the re-broadcasts and diversity-combine to decode-and forward the packet and the next set of nodes repeat this.

47. Design Approach (Cont.) – Cooperative Transport • The increased SNR at each stage is “transferred over” to the next stage. At each stage, the increased SNR can be used to select higher-level modulation schemes to increase capacity, or lower the transmit power to save energy. • The extension of the relay-oriented physical layer to accommodate cooperative diversity may be done as follows: instead of just one node relaying, multiple nodes may relay, and the “delay” block is adjusted to adequately synchronize the transmissions. The RX module now contains the technology required to diversity-combine the simultaneous transmissions.

48. Research Challenges • Creating a low-cost MANET-specific transceiver that is built for relaying from the ground up. • Determining optimal segment lengths. This includes considering error characteristics, other flows etc. Moreover, Enabling the “crossing” of segments instead of terminating them at sentinels. This might, for instance, be done using orthogonal coding techniques or opportunistically using gaps between packets. • Integrating and jointly optimizing routing and cooperative transport. This involves selecting the optimal set of nodes in each stage that will cooperatively transmit so that path capacity is maximized, energy is minimized.

49. To Sum Up • RAIN is motivated by contention control for packet-switched asynchronous wireless networks with off-the-shelf half-duplex wireless physical layers. Important architectural functions, i.e., contention and congestion control, are implemented based on wireless datalink layer mechanisms that are usually mandatory in off-the-shelf CSMA/CA wireless transceivers. • Ramanathan proposes a new MANET architecture. Its main goal is to streamline packet forwarding at intermediate wireless nodes using the future full-duplex, relay-oriented wireless physical layer without having the upper MAC or network layer involved, similar to circuit switching in synchronous networks.

50. Outline • RAIN: A Reliable Wireless Network Architecture • Challenges: A Radically New Architecture for Next Generation Mobile Ad Hoc Networks • A Cautionary Perspective on Cross-layer Design