220 likes | 615 Views
Real-Time Scheduling for WirelessHART Networks. Abusayeed Saifullah, You Xu, Chenyang Lu, Yixin Chen Department of Computer Science & Engineering. Motivation. Process control. Challenges in process control Harsh environment Real-time and reliability requirements WirelessHART
E N D
Real-Time Scheduling for WirelessHART Networks Abusayeed Saifullah, You Xu, Chenyang Lu, Yixin Chen Department of Computer Science & Engineering
Motivation Process control • Challenges in process control • Harsh environment • Real-time and reliability requirements • WirelessHART • Open standard for process industries • Based on IEEE 802.15.4 radio • Commercial deployments • Open problem for WirelessHART networks • Transmission scheduling of real-time flows for feedback control loops (sensor-controller-actuator) Controller
Outline • The real-time scheduling problem in WirelessHART • Necessary condition for schedulability • Optimal scheduling algorithm using branch and bound • Conflict-aware Least Laxity First (C-LLF) scheduling • Simulations using testbed topologies and random networks
WirelessHART Network Model • Components • A gateway, field devices (sensors and actuators) • A network manager: creates and distributes the schedule • TDMA • Spectrum diversity • Multi-channel • No spatial reuse of a same channel • Route diversity • Two transmissions conflict if they involve a common node 2 1 4 5 3 a 1& 5 are not conflict-free 4 & 5 are not conflict-free 3 & 4 are conflict-free 4
Problem • Real-time flows F={F1, F2,… , FN} • Each flow Fi has • Period Pi • Source, destination, a set of routesthrough the gateway • Deadline Di (≤ Pi) • Objective: Schedule transmissions to meet all deadlines • Optimal Algorithm: can schedule if a feasible schedule exists • NP-Hardness: Reduction from edge-coloring
Intuition for Necessary Condition • Analogous to multiprocessor scheduling • m channels ≈ m processors • But different! • Transmission conflicts play a major role in communication delay • Approach is analogous to time demand analysis • But must incorporate transmission conflicts
Necessary Condition • Necessary conditionfor schedulability • Used to prune the search space for an optimal algorithm • Provides the key insights for heuristics • For all packets released within the hyper-period • Bound transmission lifetimes • Identify the time windows analogous to busy periods • Identify the number of transmissions and conflicts in every window • Verify if the time windows are long enough to accommodate the transmissions
Transmission Lifetime y τk postk-= 3 dk=D−postk =21-3=18 • Deadline of transmission τk: dk=D−postk • Release time of τk: rk=R+prek,s+max(s−R,0) • Lifetime of τk: window [rk, dk] x b τj destination a source Transmissions for a packet with R = slot 10, D=slot 21 • D: deadline of the corresponding packet • R: Release time of the packet • prek,s: number of preceding transmissions • of τk at time slot s • postk: number of transmissions after τk Let only τj has been scheduled before slot 15 At slot 15, prek,15 =2 rk=10+2+max(15-10, 0)=17
Necessary Condition • At any time slot s, lifetime [rk, dk] of τk implies that at least • 1 transmission must happen in window [rk, dk] • 2 transmissions must happen in [rk−1, dk], if prek,s > 0 • 2 transmissions must happen in [rk, dk+1], if postk > 0 • 3 transmissions must happen in [rk − 1, dk + 1], if prek,s>0, postk > 0 • Similar analysis for all pending transmissions • Gives a set of time windows W • For every window [t1, t2] in W, sum of the number of transmissions whose lifetimes are contained in window [t1, t2] is the minimum number of transmissions that must be scheduled in [t1, t2]
Time Demand within [t1, t2] • Q1,2: smallest set of transmissions that must occur in [t1, t2] • Ψ1,2: largest set of mutually conflicting transmissions in Q1,2 • Time demand • Using m channels, at least time slots are required to schedule Q1,2 transmissions • At least |Ψ1,2| time slots are required to schedule the mutually conflicting transmissions At least max (Ψ1,2, ) time slots are required
Necessary Condition Min {(t2-t1+1)–max (Ψ1,2, ) | [t1,t2] in W} ≥ 0 Every time window in W must be able to accommodate the transmissions whose lifetimes are contained in it
Branch-and-Bound Scheduling • Uses the necessary condition to discard unnecessary branches in the search tree • Any node that cannot satisfy the necessary condition is closed • For every scheduling option in the next slot at an open node, we create a child node • Continue until we find a feasible schedule, or until there is no unexpanded node that can satisfy necessary condition
Conflict-aware Least Laxity First (C-LLF) • Intuition: enforcing the necessary condition • Conflict-aware laxity of a transmission: determined by the length of time window in which the transmission must happen and the potential conflicts in that window • Transmissions with intersecting lifetimes are the potential sources of conflict
C-LLF • Conflict-aware laxity λof transmission uv at slot s • A= set of deadlines of transmissions that involve node u and whose lifetimes intersect with the lifetime of uv • σs,d=number of transmissions that involve node u in window [s, d] for d in A λ=min{ (d-s+1) - σs,d | d in A} window size conflict Conflict-aware laxity • C-LLF • At every time slot, calculate λ for every ready transmission • Schedule the transmission with the smallest λ • Complexity: pseudo-polynomial (in terms of period)
C-LLF: An Example g • 2 flows of same period • F1(c to d): deadline 6 • F2 (e to b): deadline 5 4 5 5 4 a b 3 6 3 • At slot 1, for transmission ca(F1): • 3 transmissions at c for 3 slots ([1,3]) • σ1,3 = 3 λ= 3-3 = 0 c d 2 e • At slot 1, for transmission ec(F2): • 1 transmission at e for 2 slots ([1,2]) • σ1,2 = 1 λ= 2-1 = 1 • At slot 1, schedule ca(F1)
Evaluation: Testbed Topology Wireless Sensor Network Testbed topology of 47 TelosB motes in Bryan Hall of Washington University
Evaluation: Schedulable Ratio Number of routes between source and destination = 1 Number of channels =8 UP (Upper bound) B&B (Branch & bound) C-LLF EDF EPD(Earliest Proportional Deadline) LLF(Least Laxity First)
Scalability with Random NetworksSchedulable Ratio UP 2 routes between every source and destination pair C-LLF EDF EPD LLF
Conclusion • WirelessHART: open wireless standard for process industries • Real-time transmission scheduling for WirelessHART • Proof of NP-hardness • Challenges and opportunities for real-time scheduling theory • Optimal algorithm based on branch and bound • Conflict-aware Least Laxity First (C-LLF) • Incorporate transmission conflicts in scheduling • Near optimal performance • Scalable in large WirelessHART networks