Towards a Real Time Communica3on Framework for Wireless Sensor Networks

Transcription

1 Towards a Real Time Communica3on Framework for Wireless Sensor Networks Chenyang Lu Department of Computer Science and Engineering

2 Applica3on challenges High data rate Low latency Priori;za;on Predictability Structural Health Monitoring Process Monitoring 2

3 Outline Conflict free transmission scheduling Op;mized for queries in sensor networks. Priority based real ;me scheduling Trade off between priori;za;on and throughput. Worst case delay analysis Bridging the gap between sensor net and real ;me scheduling theory. Other projects 3

4 Query model SELECT acceleration FROM accelerators SAMPLE RATE 10Hz DEADLINE 0.1s Query periodic data collec;on from a set of sensors Query instance instance of query in a sampling period 4

5 Network model Interference Communica;on Graph Communica;on edge (AB): A s transmission may be received by B Interference edges (AD): D cannot receive when A transmits, even though D cannot decode A s transmission A B Transmissions AB and CD are conflict free if: AD and CB are not present in the graph C D Can be constructed using the RID protocol [Zhou 2005] 5

6 Overview: query scheduling Planner: Offline or when a query arrives Construct a schedule for a single query instance Reduce query latency based on transmission dependency Scheduler: Run ;me Dynamically schedule mul)ple concurrent query instances Improve throughput Maintain conflict free Priority based real ;me schedulers 6

7 Planner Plan: a sequence of steps Schedule for a single query instance Independent of other query instances A step includes a set of conflict free transmissions Respect inter transmission dependencies incurred by rou;ng or aggrega;on. Planning algorithm Construct a reversed plan Assign priori;es to nodes based on (depth, numbers of children, IDs). Assign a transmission to the node with the highest priority to the current step, if no conflict with previous transmissions assigned to the same step. Reverse the plan 7

8 Example of a plan t senders z w n l m o e d 2 a j b c f k h g s r s t e p s 0 q 1 n p 2 f k o z s 3 e l h t r 4 c j g w 5 b m 6 d p 1 q 0 8

9 Concurrent instances conflicts Slots: Reduced 1 throughput! Q1: Q2: conflicts? 9

10 Minimum step distance Minimum step distance is the smallest Δ such that: if the distance between any two steps Δ conflict free Scheduler: Enforce a gap of Δ between instances => conflict free Min. step distance: Δ = Conflict table: Δ Δ Δ Δ Δ Δ Δ 1 Δ Δ Δ Δ Δ Δ Δ 2 Δ Δ Δ Δ Δ Δ Δ 3 Δ Δ Δ Δ Δ Δ Δ 4 Δ Δ Δ Δ Δ Δ Δ 5 Δ Δ Δ Δ Δ Δ Δ 6 Δ Δ Δ Δ Δ Δ Δ 10

11 NS2 simula3on: throughput 58.6% 11

12 NS2 simula3on: latency 74.8% 12

13 Real 3me query scheduling Trade off between priori3za3on and throughput Non preemp;ve Query Scheduler (NQS) high throughput, priority inversion Preemp;ve Query Scheduler (PQS) lower throughput, no priority inversion Slack stealing Query Scheduler (SQS) uses preemp;on only when necessary Improve throughput without missing deadlines 13

14 Nonpreemp3ve Query Scheduler (NQS) Order pending queries based on priority Enforce Δ between the start )mes of consecu;ve instances High priority instance does not preempt low priority instances that have already started Plan length = 15, Δ = 8 Suffers from priority inversion

15 Preemp3ve Query Scheduler (PQS) High priority instance preempts low priority instance if they would conflict. 15

16 NQS/PQS Comparison Prioritization without preemption Prioritization with preemption Higher priority query has lower latency 16

17 NQS/PQS Comparison Prioritization without preemption Prioritization with preemption Lower query throughput 17

18 Slack stealing query scheduling (SQS) SQS combines benefits of NQS and PQS Use preemp;on only when necessary to meet deadlines Improve throughput while mee;ng all deadlines Slack the maximum ;me a query instance may be delayed without missing its deadline SQS scheduling algorithm If it has enough slack, a higher priority instance allows a lower priority instance to complete its first Δ steps Otherwise, the higher priority instance preempts the low priority instance immediately 18 0

19 NS2 simula3on: Priori3za3on NQS PQS

20 NS2 simula3on: SQS SQS

21 Worst case delay analysis Assump;on: period <= deadline Map to Response Time Analysis in real ;me scheduling theory Execu;on ;me: length of the plan L Interference from a high priority instance: Δ Blocking ;me: Δ 1 Response ;me:

22 Worst case delay analysis Assump;on: period <= deadline Map to Response Time Analysis in real ;me scheduling theory Execu;on ;me: length of the plan L Interference from a high priority instance: Δ Blocking ;me: Δ 1 Response ;me: execution time

23 Worst case delay analysis Assump;on: period <= deadline Map to Response Time Analysis in real ;me scheduling theory Execu;on ;me: length of the plan L Interference from a high priority instance: Δ Blocking ;me: Δ 1 Response ;me: worst case delay before instance l starts

24 Worst case delay analysis Assump;on: period <= deadline Map to Response Time Analysis in real ;me scheduling theory Execu;on ;me: length of the plan L Interference from a high priority instance: Δ Blocking ;me: Δ 1 Response ;me:

25 Worst case delay analysis Assump;on: period <= deadline Map to Response Time Analysis in real ;me scheduling theory Execu;on ;me: length of the plan L Interference from a high priority instance: Δ Blocking ;me: Δ 1 Response ;me: blocking time

26 Worst case delay analysis Assump;on: period <= deadline Map to Response Time Analysis in real ;me scheduling theory Execu;on ;me: length of the plan L Interference from a high priority instance: Δ Blocking ;me: Δ 1 Response ;me: interference

27 Conclusions Conflict free transmission scheduling Op;mized for queries in sensor networks. Adap;ve to workload changes. Priority based real ;me schedulers Trade off between priori;za;on and throughput. Worst case delay analysis Bridging the gap between sensor net and real ;me scheduling theory. O. Chipara, C. Lu, J.A. Stankovic, Dynamic Conflict free Query Scheduling for Wireless Sensor Networks, ICNP 06. O. Chipara, C. Lu, G. C. Roman, Real ;me Query Scheduling for Wireless Sensor Networks, RTSS

28 MLA: MAC Layer Architecture Separa;on of power management from radio core [IPSN 07] Components for sleep scheduling protocols [SenSys 07] Reusable ease development and maintenance of MAC protocols Plaporm independent reduce por;ng effort Power Management Timers Radio Core 28

29 Reusability of Components 29 B-MAC X-MAC SCP-Wustl Pure-TDMA SS-TDMA Channel Poller LPL Listener Preamble Sender Time Synchronization TDMA Slot Handler CSMA Slot Handler Low Level Dispatcher Async I/O Adapter Alarm Local Time Radio Core Other Components Reused Components

30 Solve MAC Problems Hard to develop new MAC protocols? Example: RI MAC (SenSys 08) built on top of MLA Hard to maintain mul;ple MAC stacks as OS evolves? Upgrading MLA for TinyOS >2.0.2 >2.1 took several hours Mul;ple MAC protocols survived upgrade without any change! Protocols not reusable across radio/processor plaporms? Supports both Telos and MicaZ TinyOS 2.1 version available from TinyOS contrib CVS K. Klues, G. Hackmann, O. Chipara, and C. Lu, A Component Based Architecture for Power Efficient Media Access Control in Wireless Sensor Networks, SenSys

31 Clinical Monitoring Wireless pulse oximeter Low power mesh network Barnes Jewish Hospital Deployment Orders of magnitude higher granularity than current prac;ce 1 reading/min vs. several manual readings/day) Enables early detec;on of clinical deteriora;on Highly reliable network (99.92%) O. Chipara, C. Brooks, S. Bhazacharya, C. Lu, R.D. Chamberlain, G. C. Roman, T.C. Bailey, Reliable Real ;me Clinical Monitoring Using Sensor Network Technology, AMIA

32 Structural Health Monitoring Co design of distributed sensor network architecture and structural engineering algorithms Successful damage localiza;on on lab structures Advantages over centralized approaches reduce latency by 88% increasing life;me by a factor of 3.4 under an hourly schedule G. Hackmann, W. Guo, G. Yan, C. Lu, S. Dyke, Cyber Physical Codesign of Distributed Structural Health Monitoring With Wireless Sensor Networks, ICCPS'10. G. Hackmann, F. Sun, N. Castaneda, C. Lu and S. Dyke, A Holis;c Approach to Decentralized Structural Damage Localiza;on Using Wireless Sensor Networks, RTSS 08.