Predictive Delay-Aware Network Selection in Data Offloading Haoran Yu, Man Hon Cheung, Longbo Huang, Jianwei Huang Network Communications and Economics Lab (NCEL) The Chinese University of Hong Kong (CUHK), Hong Kong Institute for Interdisciplinary Information Sciences (IIIS) Tsinghua University (THU), China Haoran Yu (CUHK) Predictive Network Selection December 2014 1 / 24
24
Embed
Predictive Delay-Aware Network Selection in Data Offloadingusers.eecs.northwestern.edu/~hyw5728/HaoranYu_home/... · Mobile Tra c Explosion Traditional ways toexpand the network capacity
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Predictive Delay-Aware Network Selectionin Data Offloading
Haoran Yu, Man Hon Cheung, Longbo Huang, Jianwei Huang
Network Communications and Economics Lab (NCEL)
The Chinese University of Hong Kong (CUHK), Hong Kong
Institute for Interdisciplinary Information Sciences (IIIS)
Type 1: User-initiated offloadingI Users decide which network (e.g., cellular or Wi-Fi) to connect
Type 2: Operator-initiated offloading (this work)I Mobile operator makes the network selection decisionI Advantages: seamless switch, optimize revenue and QoE
Multiple networks, locations, and usersI Network availability is location-dependentI Users randomly move across the locations with random traffic arrivals
System settingsI Slotted system, t ∈ {0, 1, 2, · · ·}I Set of locations, S = {1, 2, · · · ,S}I Set of users, L = {1, 2, · · · , L}I Set of networks, N = {1, 2, · · · ,N}I Location-dependent availability: Ns ⊆ N , networks available at s ∈ S
System randomnessI User l ∈ L’s traffic arrival at t, Al (t)I User l ∈ L’s location at t, Sl (t)
Operator’s online decisionI Network selection for l at t, αl (t)I Determine: User l ’s transmission rate at t, rl (α (t))I Determine: Total operation cost at t, c (α (t))
At time t, operator only observesI Traffic arrivals, A (t) (random variable)I Users’ locations, S (t) (random variable)I Data queues, Q (t) (determined by historical arrival and transmission)
Propose DNS algorithm to make online decision on α (t)
Under mild assumption on capacity region, for i.i.d. randomness:
cDNSav , lim sup
t→∞1t
t−1∑τ=0
E{c (α (τ))} ≤ c∗av + BV ,
QDNSav , lim sup
t→∞1t
t−1∑τ=0
L∑l=1
E{Ql (τ)} ≤ B+Vcmaxη .
[O (1/V ) ,O (V )] cost-delay tradeoff (V : control parameter)I Time average operation cost is within O (1/V ) of the optimalityI Time average traffic delay is bounded by O (V )I Conclusion: The operation cost can be pushed arbitrarily close to the
optimal value, but at the expense of an increase in the traffic delay
DNS has similar cost-delay tradeoff under Markovian randomness
As control parameter V increases,I Operation costs (of DNS or GP-DNS) approach the minimum valueI Queue lengths or traffic delay (of DNS or GP-DNS) become larger