A First Look at Unstable Mobility Management in Cellular Networks - PowerPoint PPT Presentation

A First Look at Unstable Mobility Management in Cellular Networks Yuanjie Li 1 Jiaqi Xu 2 Chunyi Peng 2 Songwu Lu 1 1 University of California, Los Angeles 2 The Ohio State University HotMobile’16

Ubiquitous Cellular Network Access Cellular Networks 7.9+ billion in 2015 Chunyi Peng HotMobile'16 2

Mobility Management (MM) via Handoff Cellular Networks § Seamless connectivity (via switching the serving cell) ¨ Each cell: limited radio coverage Chunyi Peng HotMobile'16 3

Desirable Handoff: Stability Why desirable? § Handoff comes at a cost 3 1 ¨ Multi-round signaling Handoff Cell 3 Cell 1 exchange ¨ Service disruption/degradation 2 4 Cell 2 movement § Converge to certain cell given an invariant setting Chunyi Peng HotMobile'16 4

Problematic Handoff Desirable Handoff 6 3 1 1 Handoff Cell 3 Cell 1 5 Cell 3 Cell 1 3 4 2 2 4 Cell 2 Cell 2 movement § Instability (persistent loop): § Stability ¨ C1->C2->C3->C1->C2->C3… ¨ Converge to certain cell Chunyi Peng HotMobile'16 5

This Work: Instability in Mobility Management - Q1: Does it exist in real networks? - Q2: Why unstable? - Q3: How to identify such risk? Caused by fundamental (persistent) conflicts in policy not by transient factors (radio dynamics etc) Chunyi Peng HotMobile'16

Q1: Does unstable MM exist in reality? - Unfortunately, Yes! Chunyi Peng HotMobile'16 7

3-Cell Loop Example § Static, 40hr-loop 1hr trace 1 Serving Cell 2 Cell 3 Cell 1 3 0 20 40 60 Cell 2 (minute) Loop every few mins Cell1: 4G Cell2: Femtocell (3G) (90% loops in 200s) Cell3: 3G 10 11 12 13 14 15 Chunyi Peng HotMobile'16 8

Negative Impacts in Real-world § Hurt both carriers and users § Excessive signaling § Performance degradation overhead (2-8x) (10+ fold slowdown) 8K 80 Downloading time (s) num. msgs per hour 76s 180s 180 Web loadtime (s) 60 150 6K 120 2.2x 40 4K 90 3.5x 60 8.5x 20 2K 12s 3s 30 5s 0 0 0 cnn 1 2 3loop 5MB file 1 2 3loop 1 2 3loop Chunyi Peng HotMobile'16 9

Q2: Why is MM unstable? Chunyi Peng HotMobile'16 10

Distributed Nature of Handoff § Each handoff: trigger-decision-execution phases § Sequence of handoffs Handoff decision@C2 Handoff decision@C1 decision decision rules rules’ … … 5 2 paras paras’ meas meas’ 3 4 1 Cell 2 Cell1 Chunyi Peng HotMobile'16 11

Handoff for Versatile Demands § Seamless connectivity § Voice/data support Handoff decision@C1 § Performance decision Each individual handoff: OK improvement rules 2 paras § Load balancing ≠ meas § … The interplay among multiples: OK C1 Chunyi Peng HotMobile'16 12

3-Cell Loop Example Rule/preference configuration@C1 (4G) § C2 (Femto) > C1 (4G) for offloading § C1 (4G) > C3 (3G) for higher-speed C1: C1 à C2 Cell 3 Cell 1 @C2 (3G Femto) § Best radio strength with same preferences for all cells Cell 2 C2: C2 à C3 Cell1: 4G @C3 (3G) Cell2: Femtocell (3G) Cell3: 3G § C1 (4G) > C3 (3G) for higher-speed C3: C3 à C1 Chunyi Peng HotMobile'16 13

From Example to Generalization § Each handoff decision: t = Fs ( s, C ) ¨ s: serving cell ¨ C: set of candidate cells ¨ Fs : decision function for serving cell s ¨ t: target cell § The sequence of handoff decisions s → Fs ( s ) → · · · c i → [ c i +1 = Fc i ( c i )] → · · · , ci ∈ C. Chunyi Peng HotMobile'16 14

From Example to Generalization § Instability = No convergence ¨ e.g., persistent loop: c → · · · c i →c i +1 → · · · c. § [Necessary stability condition] there exists at least one t, s.t. ∃ t ∈ C, t = Ft ( t, C ) § [Necessary and sufficient condition] (1) ∃ t ∈ C, t = Ft ( t, C ); (2) there exists a handoff path from the initial cell s to the desirable t Chunyi Peng HotMobile'16 15

Q3: How to detect possible instability? Chunyi Peng HotMobile'16 16

MMDIAG § In-device diagnosis ¨ Carriers: reluctant to provide network-side MM info § Two-phase: analyzer and validation static Config. MM Collector Automata Counter examples dynamic Instability Empirical Scenario Analyzer Validation Emulator Chunyi Peng HotMobile'16 17

MMDIAG § Model based on 3GPP spec § Decision logic, configuration parameters and runtime observation (scenario) static Config. MM Collector Automata Counter examples dynamic Instability Empirical Scenario Analyzer Validation Emulator Chunyi Peng HotMobile'16 18

§ Model based on 3GPP spec MMDIAG § Decision logic, configuration parameters and runtime observation (scenario) § Violation check static Config. MM Collector Automata Counter examples dynamic Instability Empirical Scenario Analyzer Validation Emulator Chunyi Peng HotMobile'16 19

MMDIAG § Scenario reconstruction and experiments § Configurations and observations in counterexample § Trace collection and comparison static Config. MM Collector Automata Counter examples dynamic Instability Empirical Scenario Analyzer Validation Emulator Chunyi Peng HotMobile'16 20

Real-World Findings - One top-tier US carrier - Los Angeles and Columbus 63 locations (outdoor) 50 spots (indoor) Chunyi Peng HotMobile'16 21

Preliminary Results L1: 4G-Femtoell-3G L2: 4G-Femtoell-2G-3G § 17 loops (idle) L3: 4G-4G § 1 loop (active) c1 (4G, band 17) c6 (3G, band 1900) c2 (4G, band 2) c5 (3G, band 850) c7 (2G) c3 (4G, band 4) c4 (Femtocell) Chunyi Peng HotMobile'16 22

Four Classes (Root Causes) L1: 4G-Femtoell-3G L2: 4G-Femtoell-2G-3G § #1: L3: 4G-4G uncoordinated handoff goals c1 (4G, band 17) c6 (3G, band 1900) ¨ 8 variants, 4G-Femto-3G c2 (4G, band 2) c5 (3G, band 850) c7 (2G) c3 (4G, band 4) c4 (Femtocell) Chunyi Peng HotMobile'16 23

Four Classes L1: 4G-Femtoell-3G L2: 4G-Femtoell-2G-3G § #2: device-side L3: 4G-4G misconfiguration ¨ 8 variants, 4G- c1 (4G, band 17) c6 (3G, band 1900) Femto-2G-3G c2 (4G, band 2) c5 (3G, band 850) c7 (2G) c3 (4G, band 4) c4 (Femtocell) Chunyi Peng HotMobile'16 24

Four Classes L1: 4G-Femtoell-3G L2: 4G-Femtoell-2G-3G § #3: Imprudent L3: 4G-4G 4G upgrade ¨ One 4G-only c1 (4G, band 17) c6 (3G, band 1900) loop § #4: uncoordinated c2 (4G, band 2) c5 (3G, band 850) c7 (2G) load balancing ¨ One 4G-only loop (active) c3 (4G, band 4) c4 (Femtocell) Chunyi Peng HotMobile'16 25

Takeaway § Largely stable in practice ¨ Instability mainly caused by Femtocells or incompatible upgrades § But in principle, instability likely exists ¨ Distributed nature ¨ Diversity and external (non-carrier) factors in case of heterogeneous networks (femtocells, small cells, WiFi, etc) Chunyi Peng HotMobile'16 26

Open Issues § Non-stability properties ¨ Handoff converges to an undesirable choice (3G/2G when 4G available) § Cooperate with network-side efforts § From detection to fix ¨ Report identified problems to carriers ¨ Assist end-devices to intervene the loop Chunyi Peng HotMobile'16 27

Summary § A first look at instability in mobility management over cellular network § Disclose real-world persistent loops caused by misconfigurations and policy conflicts § Propose MMDIAG to detect unstable MM § Call for more attention and efforts Chunyi Peng HotMobile'16 28

Thank you! Questions? Chunyi Peng HotMobile'16 29