Is IEEE TSF Scalable? L. Huang, T.H. Lai, On the scalability of IEEE ad hoc networks, MobiHoc 2002.
IEEE : how large can it be? Bandwidth: Up to 54 Mbps Good for a few hundred nodes Timing Synchronization Function Not scalable How to fix it?
802.11’s Time Sync Function (I) Time divided into beacon intervals, each containing a beacon generation window. Each station: waits for a random number of slots; transmits a beacon (if no one else has done so). Beacon: several slots in length. window beacon interval
802.11’s Time Sync Function (II) Beacon contains a timestamp. On receiving a beacon, STA adopts beacon’s timing if T(beacon) > T(STA). Clocks move only forward. faster adopts 12:0112:00 slower not adopts 12:01 12:02 12:01
Problems with ’s TSF Faster clocks synchronize slower clocks. Equal opportunity for nodes to generate beacons. 1:10 1:11 1:12 1:13 1:14 1:15 1:13 1:14 1:15 1:16 1:17 1:18 1:19 1:21 1:23 1:18 1:19 1:21 1: :21 1:22 1:23 1:25 1:28 1:31 1:23 1:25 1:28 1:31
The Out-of-Sync Problem When number of stations increases Fastest station sends beacons less frequently Stations out of synchronization
Two Types of Out-of-Sync Fastest-station out-of-sync – fastest station is out of sync with all others. k-global out-of-sync – k percent of the n(n-1)/2 links/pairs are out of sync. Questions: How often? For how long?
Fastest-station out-of-sync (1) Clock1 and Clock2: two fastest clocks d = their difference in accuracy T = length of beacon interval (0.1 sec.) Clock drift: d*T per beacon interval. If there is no beacon from fastest station in /(d*T) intervals, fastest-station out of sync occur. T
Fastest-station out-of-sync (2) How often may it occur? Once occurs, how long may it last? H = # beacon intervals with F.S. out-of-sync. L = # beacon intervals between async periods. E(H) = ? E(L)? LH
Fastest-station out-of-sync (3) n = number of stations. W + 1 = size of beacon window. P = P(n,w) = prob(fastest station wins beacon contention) W w
p = P(n,W) = ? k w 0 P(n,W,k) = prob(F.S. succeeds | it sends at slot k)
P(n,w,k) = prob(F.S. succeeds | it sends at k) = ? k w k wi i+b-1 b 0 x≥2 y slot #stations n-x-y n-1 stations 0 x≥2x≥2 #stations n-x-y y P(n-x-y, w-i-b, k-i-b)
P(n,W,k) = ?
E(H) = ? H = # beacon intervals with F.S. out-of-sync. LH
E(L) = ? e i : F.S. sends another beacon after i intervals. LH i τ = /(d*T)
E(L) = ?
Prob(Fastest station sends a beacon)
How often does fastest-node get out of sync with others?
Percentage of time fastest station out of sync with all others a 54 Mbps ∆ = 224 s d = 0.003%
How often does 25%-async occur?
Percentage of time with 25 percent of links out-of-sync a 54 Mbps ∆ = 224 s d = 0.01%
How to fix it? Desired properties: simple, efficient, and compatible with current TSF. Causes of out-of-sync Unidirectional clocks Equal beacon opportunity Single beacon per interval Beacon contention (collision)
Improve fastest station’s chance Let the fastest station contend for beacon generation more frequently than others.
Adaptive Clock Sync Protocol Station x participates in beacon contention once every C(x) intervals. Initially, C(x) =1. Always, 1 < C(x) < Cmax. Dynamically adjust C(x): x faster C(x) +1 x slower C(x) -1
Once the protocol converges Fastest station, C(x) =1 Other stations, C(x) = Cmax (Cmax= ?)
What if the fastest node leaves the IBSS? The previously second fastest now becomes the fastest. Its C(x) will decrease to 1.
What if a new fastest node enters the IBSS? The previously fastest now no longer the fastest. Its C(x) will increase to Cmax.
Compatible with current TSF Suppose some nodes do not implement the new protocol.
Performance Performance of TSFPerformance of TSF ATSP ATSP.pdfATSP.pdf
Performance of TSF
Performance of ATSP
Summary Showed: the IEEE Timing Sync Function (TSF) is not scalable. Proposed: a simple remedy compatible with the current TFS.
What’s Next? It’s 25 μs, stupid! How to deal with MANET? transmission range