1. Introduction to MS-Aloha R. Scopigno, Networking Lab – scopigno@ismb.it www.ms-aloha.eu 1

2. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions It works under mobility It simulatively overtakes CSMA/CA

3. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions

4. Based on reservation Aimed at achieving determinism Completely distributed Infrastructure would be a request too strong Dynamic clustering and master election would not scale It requires too much time and reacts slowly: not compatible with MAC needs Preventing hidden terminal issue Frequent in urban area Supporting priority (preemption) for emergency messages Blocking must be prevented for such messages Efficiently support target length messages and typical frequency In this study fixed at 200B, 10 Hz Requirements for slotted Vanets

5. Each node who has obtained a slot appends to the slot its view of all the slots (FI) Against hidden station and to enable collision detection Potentially dangerous overhead Contention Phase (slot reservation) A node starts competing for slot assignment listening to Slot (free busy) N FIs coming from its neighbors The node transmits a data packet into a slot considered idle, together with its FIs MS-Aloha Base Mechanisms (i)

6. The reservation of a slot is performed through two distinct phases The slot reservation through the FI True slot occupation In the period between slot(K) and slot(K+N) the channel is monitored to detect any reservation Check on slot and by FI analysis When slotK begins, the node transmits its packet if it still has the reservation. Continuous monitoring to face mobility MS-Aloha Base Mechanisms (ii)

7. Slot: channel time space dedicated to a single host for data transmission. N: number of slots within a single frame. FI: (Frame Information): Structure containing information about the status of each slot. Required to prevent hidden station In this presentation: Same Physical Layer of 802.11p (12Mbps, 10Mhz ch @5.9GHz, QAM16-1) Frame: 100ms (10Hz application Rate) Payload: 200 Bytes If FI=12 bits per slot and Tg: 1 us, then 224 slots (of 446 us) Other setting (e.g. relaxed guard time) in other studies available in www.ms-aloha.eu MS-Aloha Format (i)

8. STI (8bit) Address1(48 bit) Address2(48 bit) Sequence Number (12bit) Fragment Number (4bit) FIbit(1bit) STI: source identification Address1: source address Address2: destination address SequenceNumber: field indicating the sequence number of each packet FragmentNumber: used in case of frame fragmentation FIbit: bit indicating the presence of the FI before the payload (sent in slot0 only) Payload: CRC: used to highlight any errors during transmission MS-Aloha Format (ii)

9. FI field FI: (Frame Information): Structure containing information about the status of each slot Each slot information is composed of: STI: the short identifier of the node PSF (Priority Status Field): field indicating the priority of data transmitted in the slot. The values ranging from 1 to 3 (growing priority). STATE: 2-bit flag indicating channel state STI (8 bits) PSF (2)State (2)

10. Time Efficiency The Issue of Overhead (i) The main concern is about the overhead implied by MS-Aloha The overhead of MS-Aloha is fixed CSMA/CA introduces a protocol overhead too, but it is variable and hard to be measured Comparison by simulations in case of unicast Both broadcast and Unicast: In Broadcast CSMA/CA does not involve backoff (no ACKs)  no real OH The side effect of collisions should be taken into account 100-200 fixed nodes on two lanes Point-to-point full duplex traffic at variable application rate Peers in distinct Lanes Inter-Node-Dist 4m; Inter-Peer-Dist 60m 37dbm TX, -85dbm RX (benefits for CSMA)

11. The Issue of Overhead (ii) Unicast (100) Inter-packet time inside a flow (Average on the 100 flows) –Time between two consecutive packets correctly received CSMA/CA saturation starts at 15Hz – variable, fixed on average, higher than MS-Aloha

12. The Issue of Overhead (iii) Unicast (200) Inter-packet time inside a flow (Average on the 100 flows) –Time between two consecutive packets correctly received CSMA/CA saturation starts at 10Hz – variable, fixed on average, higher than MS-Aloha

13. Inter-packet time inside a flow (Average on the 100 flows) –Time between two consecutive packets correctly received CSMA/CA saturation starts at 15Hz – variable, fixed on average, higher than MS-Aloha The Issue of Overhead (iv) Broadcast (100)

14. Inter-packet time inside a flow (Average on the 100 flows) –Time between two consecutive packets correctly received CSMA/CA saturation starts at <10Hz – variable, fixed on average, higher than MS-Aloha The Issue of Overhead (v) Broadcast (200)

15. MS-Aloha (224 slots, 200B Appl.Layer, 12Mbps) 446  s per slot (including guard-time) Payload_Time = 200*8/12Mbps = 133  s Overhead_Time= 313  s (3.756 bit_time @ 12Mbps) Overhead/Payload = 2,35 η = 1/(1+2.35) ≈ 0,3 (including Ethernet-like Overhead) CSMA/CA (200B Appl.Layer, 12Mbps) 8-50 Hz Appl. Rate From interpacket time inside a flow to interpacket time in the air 1.000-3.500  s IPT unicast; 500-5.000  s IPT broadcast Payload_Time = 200*8/12Mbps = 133  s Overhead_Time= 867-3.367  s unicast; 367-4.867  s broadcast Overhead/Payload = 6.5-25 unicast; 2.7-36 broadcast η = 1/(1+{OH}) ≈ 0,13  0.04 unicast; 0,27  0.03 broadcast The Issue of Overhead (vi)

16. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions It works under mobility It simulatively overtakes CSMA/CA

17. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions It works under mobility

18. TDMA algorithms are usually for fixed or slowly varying topologies Fixed networks (RR-Aloha) or free-space (line-of-sight), low density and slowly varying mutual positions (STDMA) Even if standard the may NOT be suitable (!) They do not fit the requirements of dynamic environments such that of Vanet A node can appear suddenly due to obstructions Hidden terminals are much more frequent than in free space The density of nodes is so high to make hidden collisions more frequent These have a direct impact on the efficiency and the quality of the services MS-Aloha solves these issues with a first set of proprietary mechanisms Mechanisms first published under the name of «RR-Aloha+ functions» Three tricks: memory refresh, signaling semantics, scalability of label space The properness of the solutions has been validated through simulations. Typical Unresolved Issues of Other Slotted Solutions

19. Simulations highlight the first simple, yet unresolved issue concern the refreshing rate for the information on channel state In case the information is not refreshed, once a slot j is assigned to node M, the slot state would be frozen The slot would be continuously announced busy also if the node gets switched-off Additionally the information would jump too many hops In a vehicular environment, the same would happen if the node M got far from the radio range of its previous neighbourhood Moreover M would announce fallacious information - based on a radio range which is not actual On each node the memory needs to be refreshed periodically Simulations involving node mobility highlight this as the primary cause of inefficient slot allocation It is shown to works if information is refreshed once per MS-Aloha Period Additionally information on slot j is refreshed when the elapsed time has reached the position j 1. Memory Refresh

20. In DTDMA and MS-Aloha the problem of hidden terminal is counteracted by message broadcasting with FI In case of fixed nodes, each node expects confirmation of slot assignements by all the nodes in its neighbourhood The assignement is result a logic AND among received Fis If the ad-hoc network is continuously changing it is hard to know what one's neighbourhood is like Not all the nodes can be required to be always connected to confirm If a new node switches on, it ‘0’ states in the FI will reset all the connections The information carried by FI is managed by a logical OR The semantic is changed: conflicting FIs - rather than acks – drive changes 2. Signaling Semantic (i)

21. If channel state are managed by AND, 1 bit is enough to describe channel state Only if all FI agree on the assignment, the busy state is confirmed If a collision is detected, it is announced just by “free” message (thanks to AND logic) In steady state it may work; with mobility and OR it gets ambiguous  example follows In order to solve this issue, the STATE subfield is extended to two bits 2 bits allows to ditinguish the following slots: free, busy and collision An additional variation in the semantic: collisions require an explicit indication Simulations show that the overhead and latences introduced by the additional bit are negligible while make the VANET stable 2. Signaling Semantic (ii)

22. Trasmission Order Slot 0: Slot 1: Slot 2: nodes receiving from Slot 3: During slot 3, node will send acknowledgement about into slot 2 of its FI FI Example: Why an Additional Bit is Required

23. Trasmission Order Slot 0: Slot 1: Slot 2: Receive from Slot 3: FI Slot 4: The node notices the collision and send slot 2 as free on its FI So the nodes sense a collision status affecting slot 2, then set it as free (Busy=0), while the nodes do not change the slot 2 status Busy = 0 In the next FIs, the nodes which have detected a collision will send slot2 status as free of its FI This way the collision notification gets missed! The remaining nodes will send an ack about slot2 assignment, without detecting properly the collision, also due to the OR operation. Example: Why an Additional Bit is Required

24. 8-bit labels STI used to identify each node inside the communication area: 256 possible values STI are used to identify what node is using each reserved node STI are used instead of MAC addresses (typically 48-bit wide numbers) to avoid excessive overheads in the FIs In urban areas the label space may be a very strong limit However the same label can be re-used in different slots The purpose of STI is collision detection - different nodes using the same slot Label+Slot  Node Identification Still statistically not-negligible event of two hidden terminals chosing the same slot and the same STI Scalability finally solved assigning STI a “temporary meaning” STI changed by the nodes directly receiving from node A into STI’ They know also A’s MAC and compute a new STI’ based on STI and MAC The nodes which do not receive from A just know STI’. The other know that STI and STI’ represent the same node A At next period the STI’ is changed by A into STI’’ and so on. Collision are, soon or later, detected 3. Scalability of STI Label Space (i)

25. 3. Scalability of STI Label Space (ii)

26. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions It works under mobility It simulatively overtakes CSMA/CA

27. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions It simulatively overtakes CSMA/CA

28. MS-Aloha Two main issues can still hinder the exploitation of MS-Aloha in a VANET scenario: The scalability of the protocol (number of available free slots) Its capacity to strongly react to changing conditions due to mobility Simulations show that the unconstrained multihop forwarding of channel state is harmful Slot reservation is extended beyond the bounds of wireless coverage Causing resource waste and slot depletion Mobility introduces a not negligible probability of getting closer to nodes which have been assigned the same slot This becomes more relevant when nodes move in opposite directions The number of collisions grows high The effect is disruptive if slot re-use is hindered Among the causes: slot state forwarding with no limitations on the number of hops

29. Limitation of FI Forwarding So far the channel-state is described by two bits (State) Only 3 states are used (free ‘00’, busy ‘10’, collision ‘01’ ) One free configuration (say ‘11’) The free configuration can be exploited to keep trace of number of hops the information is forwarded over When some information on slot reservation is not directly detected, it is announced as 2-hop (’11’) Nodes which receive it they know that they should not use the slot but should not forward this information either This solutions have been demonstrated, by simualtions, to: Decrease the logical radius of propagation of a slot reservation Improve of resource re-use. Busy 2-HopFree

30. Improving Slot Re-Use (i) Slot re-use can be further improved setting a higher threshold on minimum reception power If the received power is lower than a given threshold THR the message IS considered for MS-Aloha but does not contribute to the FI messages It conceptually corresponds to lowering the radius of cluster of nodes which perceive a slot x assigned to a node A Instead, acting on the transmitted power would affect the SNR Further improvement by introducing a mechanism which regulates the THR dynamically THR defined on each node separately based on its perception Blocking completely prevented Simulations show that it works Effects on slot reuse (increased) Effects on Packet Delivery Rate (PDR) Lowered at higher distances but kept high close to the transmitter

31. Improving Slot Re-Use (ii) Slot Reuse: -96 dbm: 1.968 -86 dbm: 2.040 -80 dbm: 2.174 Sent Packets: CSMA/CA: 100% (*) MS-Aloha -96: 92,50% MS-Aloha -86: 94,75% MS-Aloha -80: 99,50% (*) far from saturation The average does not change much –Slot re-use is also a statistical event: the point is to make it possible However potentially still scalable –Less blocked nodes and for less time –More unused slot

32. Simulation: Settings The MS-Aloha has been implemented on NS-2 Most simulations use MS-Aloha set as follows each slot lasts 0.447 ms 224 slot per frame (the overall frame takes about 0.1sec) Packet generation rate of 10Hz Also other settings adopted 200 slots and over 78.5 µs guard time – relaxed synchronization The simulation adopts the following scenarios: Simulation lasts 2000 sec. Nakagami model was used to model propagation and urban grid with corner obstruction (extra attenuation) Transmitted power 7dbm or 20 dbm Wireless reception sensitivity -96dbm 400-900 nodes (speed in the range 50-120Km) Circular topology (radius R=1Km) with four lanes or Urban topology with grid 150m blocks and 750m-wide map In all the simulations MS-Aloha performs better than (or as well as) CSMA/CA in terms of PDR and determinism

33. In order to quantify results the following metrics adopted PDR (Packet Delivery Rate): function that shows how much a node is likely to receive a packet varying the distance from the transmitting node; Suitable for both MS-Aloha and CSMA/CA In CSMA/CA it is affected with high congestion Mean Collisions: the average number of collisions detected on the same slot, over the whole simulation and all the nodes; Suitable for only MS-Aloha Slot Re-Use: number of times a slot is re-used by different nodes (at a given time). Suitable for aloha MS-Aloha See previous slides Determinism is hardly measured but it is Close to 100% for MS-Aloha (fixed delays and high PDR, only affected by slot collisions) Lower in CSMA/CA, due to unpredictable delay (non-deterministic transmission time due to collision avoidance) and lower PDR (non- deterministic reception) Simulation: Metrics

34. Simulations: PDR Whatever the threshold MS-Aloha achives a higher PDR than CSMA/CA and a negligible worsening where reception is already low Higher thresholds in MS-Aloha force slot re-use which cause interferences and worsen PDR but quite far fram the transmitter: MS- Aloha preserves time/space- coordination

35. Simulations: Collisions Simulations: Collisions Multi-hop FI forwarding vs 2-hop Number of Collisions MultiHop Number of Collisions 2-Hop Only collisions due to mobility!

36. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions It works under mobility It simulatively overtakes CSMA/CA

37. Introduction: Concepts and Figures Introduction: Concepts and Figures First Proprietary Mechanisms: RR-Aloha+ First Proprietary Mechanisms: RR-Aloha+ Proposed Extensions Proposed Extensions Simulative Settings Simulative Settings The Final Version: MS-Aloha The Final Version: MS-Aloha Proposed Extensions for Scalability Proposed Extensions for Scalability RR-Aloha+ & MS-Aloha Simulations RR-Aloha+ & MS-Aloha Simulations Preemption and Conclusions Preemption and Conclusions

38. Preemption (i) Preemption as an additional solution against channel blocking Acting on service differentiation and aimed at QoS guarantee Each station accesses the channel with a priority, variable in [1-4] (2 bits) The priority is announced in a subfield of the FI field Whenever a node with higher priority needs to transmist, it can override a node with lower priority E.g. Node 1 can transmit in slot 5 even if it is already occupied by node 2, if node 2 has a lower priority In a possible practical scenario nodes have the highest priority only for emergency messages Normal access uses 3 lower classes E.g.:1-emergency; 2-channel-access; 3-assistance, 4-entertainment

39. Preemption (ii) Questions to be answered: Can preemption help saturate the channel? Does preemption work also under saturation? Can it really gain channel access Several simulations. Following results achieved with: 858 nodes, average speed 80km, TX power: 2 dbm 5x5 grid (150m distance); 2-lane roads Application rate at 30Hz With and without preemption With preemption each nodes tries to have a High-Priority slot and a Low-Priority slot Results (2.000 sec of simulated time) Transmitted packets: with preemption-34.360; w/o preemption-18.028; With preemption: HP packets: 17.980; LP packets 16.378

40. Preemption (iii) Collisions without preemption Collisions at slot0 with preemption

41. Thank You for Your Kind Attention R. Scopigno, Networking Lab – scopigno@ismb.it www.ms-aloha.eu 41