Experimental Evaluation of Large Scale WiFi Multicast Rate Control, By: Varun Guptay, Craig Guttermany, Yigal Bejerano, Gil Zussmany
- 1. Experimental Evaluation of Large Scale WiFi Multicast Rate Control Presented By: Belal Essam ElDiwany Varun Guptay, Craig Guttermany, Yigal Bejerano, Gil Zussmany 2 January 2017 1/20
- 2. | Agenda • Motivation • Related Work • Objective • MuDRA Algorithm • Experimental Evaluation • Performance Comparison 2 January 2017 2/20
- 3. |Motivation • As a solution for multimedia delivery in crowded areas, WiFi multicast to very large groups has gained attention considerably. • So far, most recently proposed schemes do not provide performance guarantees and none have been tested at scale. • To address the issue of providing high multicast throughput with performance guarantees, authors present the design and experimental evaluation of the Multicast Dynamic Rate Adaptation (MuDRA) algorithm. 2 January 2017 3/20
- 4. |Motivation Cont’d • Experimental evaluation of MuDRA on the ORBIT testbed with over 150 nodes shows that MuDRA outperforms other schemes and supports high throughput multicast flows to hundreds of receivers while meeting quality requirements. 2 January 2017 4/20
- 5. |Related Work • Several solutions were proposed for Multimedia (e.g., video) delivery over crowded venues. • Most of them are based on dense deployments of Access Points (APs) and require considerable capital and operational costs, may suffer from interference between APs, and others. • Multicast offers another approach for video delivery to large groups of users interested in venue specific content (e.g., sports arenas, entertainment centers, and lecture halls). 2 January 2017 5/20
- 6. |Related Work Cont’d • However, WiFi networks provide limited multicast support at low rates without a feedback mechanism that guarantees service quality. • To improve performance, there is a need for a multicast system that dynamically adapts the transmission rate. • A major challenge in designing Multicast Rate Adaptation (RA) system is to: • Obtain accurate quality reports with low overhead (i.e., a multicast system should conduct efficient RA based on only limited reports from the nodes). • Note: • Packet delivery ratio (PDR), is the ratio of the received packets to the transmitted ones, a QoE (QoS) metric. • A normal node is the one having its PDR above L (PDR threshold). • Otherwise, it is considered as an abnormal node. 2 January 2017 6/20
- 7. |Objective • To develop a practical and efficient rate control system which satisfies the following requirements: (R1) High throughput – Operate at the highest possible rate, termed as the target rate, while preserving SLAs. (R2) Service Level Agreements (SLAs) – Given L , and a Population-Threshold X, the selected rate should guarantee that at least X% of the nodes experience PDR above L (i.e., are normal nodes). • This provides an upper bound of Amax = n. (1-X) on the number of permitted abnormal nodes. (R3) Scalability – Support hundreds of nodes. (R4) Stability – Avoid rate changes due to sporadic channel condition changes. (R5) Fast Convergence – Converge fast to the target rate after long-lasting changes (e.g., user mobility or network changes). (R6) Standard and Technology Compliance – No change to the IEEE 802.11 standard or operating system of the nodes. 2 January 2017 7/20
- 8. |The Multicast Dynamic Rate Adaptation (MuDRA) algorithm • Intuitive Note: multicast packets are not acknowledged. • The overall MuDRA algorithm relies on three main components: • (i) Feedback (FB) Node Selection. • (ii) Rate Decision (Procedure 1): Utilizes the FB reports to determine the highest possible rate, termed the target-rate. • (iii) Stability Preserving Method (Procedure 2): A window based method that maintains rate stability in the event of sporadic interference and after an RA decision. 2 January 2017 8/20
- 9. |A. Feedback (FB) Node Selection • The FB node selection process, termed “K- Worst”, where the AP selects K nodes with the worst channel conditions as FB nodes (the node’s channel condition is determined by its PDR). 2 January 2017 9/20
- 10. |B. Rate Decision Rules and Procedure • Introducing the target rate: In the experiment, the target rate is 36Mbps which is the highest rate above which the SLA requirements will be violated.
- 11. |B. Rate Decision Rules and Procedure Cont’d • Let At and Mt denote the number of abnormal and mid-PDR nodes at time t, respectively. Authors obtain the following empirical property. • Property 1 (Target Condition): Assume that at a given time t, the following condition holds, then almost surely, the AP transmits on the target-rate at time t.
- 12. |B. Rate Decision Rules and Procedure Cont’d • The rate changing rules are as follows:
- 13. |B. Rate Decision Rules and Procedure Cont’d • The rate change process is based on the aforementioned rules is as follows:
- 14. |C. The Stability Preserving Method cont’d • It is desirable to change the rate as soon as Rules I or III are satisfied to minimize QoE disruption. • Authors observed that such a strategy can cause severe rate and throughput fluctuations. • To address this, authors introduce in Procedure 2, a process which balances fast convergence with stability. 2 January 2017 14/20
- 15. |MuDRA Algorithm
- 16. |Reporting Interval Duration • For immediate response to changes in service quality, the status reports should be sent as frequently as possible, (i.e., minimal reporting interval). • The control overhead comprises of unicast FB data sent by nodes and multicast data sent by AP to manage K FB nodes.
- 17. |Reporting Interval Duration cont’d • With proper calculation for Appropriate Reporting Interval Duration T, authors get • Where ΔPDR : reduction in PDR. T : the reporting interval K : upper bound on FB nodes D : the TX duration of multicast msg d : the TX duration of FB msg Therefore, design your T based on your acceptable value for ΔPDR
- 18. |Experimental Evaluation of MuDRA • MuDRA’s operation over 300 seconds with 162 nodes: • (a): The population of abnormal nodes (2-3 most of the time). • (b): The AP converges to the target rate after the initial interference spike in abnormal nodes at 15s. The AP successfully ignored the interference spikes at time instants of 210, 240, and 280s to maintain a stable rate. • (c): The overall control overhead.
- 19. |Performance Comparison • Background traffic: two nodes near the center of the grid that exchange unicast traffic at a fixed rate of 6Mbps in a periodic on/off pattern with on and off periods 20s each. • MuDRA achieves 2x higher throughput than pseudo-multicast while sacrificing PDR only at a few poorly performing nodes. • While the fixed rate and SRA schemes can obtain similar throughput as MuDRA, they do not meet the SLA requirements.
- 21. Backup: The Experiment • In all the experiments, one corner node served as a single multicast AP. The other nodes were multicast receivers. • The AP used 802.11a to send a multicast UDP flow, where each packet was 1400 bytes. • The AP used the lowest supported transmission power of 1mW = 0dBm to ensure that the channel conditions of some nodes are marginal. 2 January 2017