Embed Size (px)
Citation preview
Real-‐Time Wireless Control Networks
Chenyang Lu Computer Science and Engineering
Outline Ø WirelessHART: a star=ng point Ø Real-‐=me scheduling for WirelessHART Ø Challenges and direc=ons of wireless control networks
2
WirelessHART Industrial wireless standard for process monitoring & control
3
4
Characteristics Ø Real-‐=me and reliable in hash industrial environments Ø Time Division Mul=ple Access (10ms slot) Ø Mul=-‐channel Ø Route diversity Ø No spa=al reuse of the same channel
Ø Centralized network manager q Collects topology informa=on from the network q Generates routes and global transmission schedule q Reconfigures when devices/links break
4
Time Synchronization Ø WirelessHART protocol
q Gateway is the root source of =me. q When a device receives a packet
• Δt = =me of arrival – expected arrival =me • sends Δt to the sender via ACK
q The sender adjusts =me.
Ø Benefits of accurate clocks q Reduce guard =me used to accommodate clock skew
ü Shorter slot
q Reduce frequency of clock sync ü Lower overhead ü Be^er scalability
5
Real-‐Time Scheduling for WirelessHART Goals Ø Real-‐=me transmission scheduling à meet end-‐to-‐end deadlines Ø Fast schedulability analysis à online admission control and adapta=on
Approach Ø Leverage real-‐=me scheduling theory for processors
Ø Incorporate wireless characteris=cs Ini=al Results Ø Dynamic priority transmission scheduling [RTSS’10] Ø Fixed priority transmission scheduling
q End-‐to-‐end delay analysis [RTAS’11] q Priority assignment [ECRTS’11]
Ø Rate selec=on for wireless control [RTAS’12]
6
Real-‐Time Scheduling for WirelessHART Goals Ø Real-‐=me transmission scheduling à meet end-‐to-‐end deadlines Ø Fast schedulability analysis à online admission control and adapta=on
Approach Ø Leverage real-‐=me scheduling theory for processors
Ø Incorporate wireless characteris=cs Ini=al Results Ø Dynamic priority transmission scheduling [RTSS’10] Ø Fixed priority transmission scheduling
q End-‐to-‐end delay analysis [RTAS’11] q Priority assignment [ECRTS’11]
Ø Rate selec=on for wireless control [RTAS’12]
7
Real-‐Time Flows Ø Flow: sensoràcontrolleràactuator over mul=-‐hops
Ø A set of flows F={F1, F2, …, FN} ordered by priori=es Ø Each flow Fi is characterized by
q A source (sensor), a des=na=on (actuator), route through the gateway (where controller is located)
q A period Pi q A deadline Di ( ≤ Pi) q Total number of transmissions Ci along the route
8
highest lowest priority
Scheduling Problem Ø Fixed priority scheduling
q Transmissions ordered based on the priori=es of their flows
Ø Flows are schedulable if Ri ≤ Di Fi F
Ø Goal: efficient end-‐to-‐end delay analysis
q Gives an upper bound of the end-‐to-‐end delay for each flow q Used for online admission control and adapta=on
9
€
∈end-to-end delay of Fi
deadline of Fi
€
∀
End-‐to-‐End Delay Analysis Ø A lower priority flow is delayed due to
q channel conten8on: when all channels are assigned to higher priority flows in a slot q conflict: its transmission and a transmission of a higher-‐priority flow involve a same node
10
2 1
1 and 5 are conflicting 4 and 5 are conflicting
4 5 3
3 and 4 are conflict-free
Ø Each type of delay is analyzed separately
Ø Combine both delays à end-‐to-‐end delay bound
Insights Ø Transmission vs. mul=processor scheduling
q Similar: channel conten=on q Different: transmission conflicts
Ø Channel conten=on à mul=processor scheduling q A channel à a processor q Flow Fi à a task with period Pi, deadline Di, execu=on =me Ci q Leverage exis=ng response =me analysis for mul=processors
Ø Account for delays due to conflict with high-‐priority flows
11
!"#$%&'"(!"#
&!"#$%&'"(!"$
)*$%+*,Ø When low priority flow Fl and high priority flow Fh, conflict, Fl is delayed
Ø Q(I,h): #transmissions of Fh sharing nodes with Fl q In the worst case, Fh can
delay Fl by Q(l,h) slots q e.g., Q(l,h) = 5 à Fh can
delay Fl by 5 slots
Delay due to ConFlict
12
Fl delayed by 2 slots
Fl delayed by 2 slots
Fl delayed by 1 slot
Tighter Bound on ConFlict Delay
13
!"#$%&'"(&!"
!"#$%&'"(&!#
)*$%+*,
€
Δ(l,h) =Q(l,h) − ∑j=1
σ ( l,h )
(δ j (l,h) − 3)
Total number of MCP of length at least 4
Length of an MCP
Ø Q(I,h) can overes=mate the delay
Ø Δ(I,h): a more precise bound of delay Fh can cause Fl
Ø Maximal common path (MCP) between two flows q Maximal overlap on their routes q On an MCP, Fl can be delayed by Fh
by at most 3 slots
Q(I,h)=8 but Δ(I,h)=3
Total Delay due to ConFlict Ø In a =me interval of t slots the delay caused by Fh on Fl is
upper bounded by
Ø The total delay of Fl due to transmission conflicts with higher priority flows is upper bounded by
14
€
Θ l (t) = ∑Fh ∈hp(Fl )
t
Ph
.Δ(l,h)
€
t
Ph
.Δ(l,h)
Ph is period of Fh
hp(Fl) is the set of higher priority flows of Fl
Complete Analysis Ø Rkch : upper bound of end-‐to-‐end delay of Fk considering that
it is delayed only due to channel conten=on
Ø Rkch,con : upper bound of end-‐to-‐end delay of Fk considering
both channel conten=on and transmission conflict
Ø For every flow in decreasing order of priority q Step 1: derive an upper bound assuming it does not conflict with any
higher priority flow. q Step 2: incorporate the conflict delay into the bound of Step 1
15
conflict channel contention
Step 1 for Flow Fk Ø Ωk(x): total delay that the higher priority flows can cause on
Fk due to channel conten=on in an interval of x slots q Determined considering the end-‐to-‐end delay Rich,con of every higher
priority flow Fi q Analyzed based on the response =me analysis for mul=processor
[Guan RTSS 2009]
Ø Rkch is the minimum value of x determined by a fixed-‐point
algorithm in equa=on
16
€
x =Ω
k(x)
m
+ C
k
m: total number of channels Number of transmissions along the route of Fk
Step 2 for Flow Fk Ø Rkch,con is the minimum value of y that solves the following
equa=on using a fixed-‐point algorithm
17
€
y = Rk
ch+ Θk (y)
Delay of Fk assuming it does not conflict with any higher priority flow
Total delay of Fk due to conflict with higher priority flows
Ø If x (in Step 1) or y (in Step 2) exceeds Dk (deadline), the algorithm terminates and reports the case as unschedulable
The analysis runs in pseudo polynomial =me.
Simulations Ø Real network topologies
q Testbed of 48 TelosB motes
Ø Random topologies
Ø Priority assignment policies q Deadline monotonic (DM) q Propor=onal Deadline monotonic
Ø Metrics q Acceptance ra=o q Pessimism ra=o
18 €
=Numberof test casesdeemed schedulableby analysis
Total test cases
€
=Analyticalupperboundof end toend delay
Worst caseend toend delayobserved in simulation
Testbed topology
40 60 80 1000.2
0.4
0.6
0.8
1
% Source or destination nodes
Acc
epta
nce
ratio
Simulation (1 route)Our analysis (1 route)Simulation (2 routes)Our analysis (2 routes)
Acceptance Ratio (Testbed Topology)
19
Ø Number of channels=12 Ø Priority assignment policy: Deadline Monotonic
Pessimism Ratio (Random Topology)
20
30 50 70 90 110 130 150 170 190 210 230 250
1
1.2
1.4
1.6
1.8
2
2.2
Pe
ssim
ism
ra
tio
Number of nodes
What we have so far… Ø Real-‐=me wireless is a reality today
q Industrial standards: WirelessHART, ISA100 q Real deployments in the field
Ø Star=ng a real-‐=me scheduling theory for wireless q Leverage real-‐=me processor scheduling q Incorporate unique wireless proper=es
Ø What’s next?
21
Wireless Dynamics Challenges Ø Wireless links change dynamically. Ø Requires global rescheduling. Ø Current approach too rigid?
Approaches Ø Flexible and dynamic scheduling
q e.g., RTQS avoids conflicts by enforcing inter-‐release =mes locally [RTSS’07]
Ø Mixed cri=cality in wireless q Sacrifice non-‐cri=cal flows when links break q Maintain guarantees to cri=cal flows
22
Scheduling-‐Control Co-‐Design Challenge Ø Wireless resource is scarce and dynamic Ø Cannot afford separa=ng scheduling and control
Approaches Ø Schedule to op=mize control objec=ves, not to meet deadlines
q e.g., rate selec=on for wireless control [RTAS’12]
Ø Achieve fault tolerance through wireless and control co-‐design Ø Scheduling for self-‐triggered and event-‐based control
23
Scalability Challenges Ø Centralized approach does not scale
q Network management q Feedback control loop
Ø WirelessHART: A gateway can support up to 80 devices
Approaches Ø Hierarchical network management Ø Local adapta=on Ø Peer-‐to-‐peer control Ø Synchronized distributed clocks as =me sources Ø Key: Scale without losing predictability!
24
Summary Ø Real-‐=me wireless is a reality today
q Industrial standards: WirelessHART, ISA100 q Real deployments in the field
Ø Star=ng a real-‐=me scheduling theory for wireless q Leverage real-‐=me processor scheduling q Incorporate unique wireless proper=es
Ø Tremendous opportuni=es ahead q Op=mize for control q Robust under wireless dynamics q Scale to 10,000+ nodes
Ø Integrate protocol design and scheduling theory
25
References Ø A. Saifullah, Y. Xu, C. Lu and Y. Chen, End-‐to-‐End Delay Analysis for Fixed
Priority Scheduling in WirelessHART Networks, RTAS 2011. Ø A. Saifullah, C. Wu, P. Tiwari, Y. Xu, Y. Fu, C. Lu and Y. Chen, Near Op=mal
Rate Selec=on for Wireless Control Systems, RTAS 2012. Ø A. Saifullah, Y. Xu, C. Lu and Y. Chen, Priority Assignment for Real-‐=me Flows
in WirelessHART Networks, ECRTS 2011. Ø A. Saifullah, Y. Xu, C. Lu and Y. Chen, Real-‐=me Scheduling for WirelessHART
Networks, RTSS 2010. Ø O. Chipara, C. Lu and G.-‐C. Roman, Real-‐=me Query Scheduling for Wireless
Sensor Networks, RTSS 2007.
h?p://wsn.cse.wustl.edu/index.php/Real-‐Time_Wireless_Control_Networks
26
