CAP for Mobility Support Yuanjie Li 1 , Zengwen Yuan 1 , Chunyi Peng 2 , Songwu Lu 1 1 University of California, Los Angeles 2 The Ohio State University
New Mobile Services in Future 5G � � � � Real-time virtual reality Remote mobile healthcare Emergency communication ed r y er � , � � High-speed mobility Mission-critical communication Safe autonomous driving 2 �
Key Enablers for 5G Mobile Services Advanced Advanced wireless technologies mobility support 3
Wide-Area Mobility via Cellular Networks Cellular Networks (e.g., 3G/4G) ... � � � � ed r y er � , 4 � � �
How does Mobility Support Work in 4G LTE? • Span on multiple network nodes • Involve multiple control procedures Internet Data-plane Control-plane Routing update Session state transfer Location update request User profile update Radio conn. setup Tracking Area 1 Tracking Area 2 Location update accept Base station Mobility controller User profile database 5 Gateway (eNB) (MME) (HSS) (SGW/PGW)
Three Expectations for Mobility Support Immediate Correct Continuous data service data service data service Once route is Forward packets Even in presence connected under right policies of failures Can we always achieve all of them? 6
This talk • A formal analysis of fundamental limits in mobility support • From the distributed system perspective • Fundamental limits: CAP theorem on generic mobility support • Rethink the tradeoff between different properties • Argument: 5G needs a paradigm shift in its mobility support 7
Outline • Q1 : What properties do we want? • Q2 : Can we achieve all properties simultaneously? • Q3 : How to balance different properties? 8
What properties do we want? From mobile user demands to formalization 9
What Properties do We Want? • Correctness : Forward every packet under correct policies Internet Data-plane Control-plane Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 10 Gateway (eNB) (MME) (HSS) (SGW/PGW)
What Properties do We Want? • Correctness : Forward every packet under correct policies • 4G LTE: correctness via sequential consistency Internet Data-plane Control-plane Routing update Session state transfer Location update request User profile update Radio conn. setup Tracking Area 1 Tracking Area 2 Location update accept Base station Mobility controller User profile database Gateway (eNB) (MME) (HSS) (SGW/PGW)
What Properties do We Want? • Correctness : Forward every packet under correct policies • 4G LTE: correctness via sequential consistency Radio conn. setup Location update request Session state transfer Routing update User profile update Location update accept Base station Mobility controller User profile database Gateway (eNB) (MME) (HSS) (SGW/PGW)
What Properties do We Want? • Correctness : Forward every packet under correct policies • Availability : start data service once route is connected Internet Data-plane Control-plane Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 13 Gateway (eNB) (MME) (HSS) (SGW/PGW)
What Properties do We Want? • Correctness : Forward every packet under correct policies • Availability : start data service once route is connected • Partition tolerance : continuous operation once route is connected Internet Data-plane Control-plane Session state transfer Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 14 Gateway (eNB) (MME) (HSS) (SGW/PGW)
Can we achieve all properties? CAP theorem on mobility support 15
CAP Theorem for Mobility For any mobility support, it is impossible to always guarantee sequential c onsistency, a vailability, and control-plane p artition tolerance simultaneously. 16
Example: Control-Plane Partition Between 4G Tracking Areas Internet Data-plane Control-plane How should I operate now? Session state transfer Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 17 Gateway (eNB) (MME) (HSS) (SGW/PGW)
Option 1 (4G LTE): Consistency Over Availability • Suspend data service until the partition is recovered • Guaranteed correctness ✓ ✓ Internet Data-plane Control-plane Session state transfer ✓ ✓ Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 18 Gateway (eNB) (MME) (HSS) (SGW/PGW)
Option 1 (4G LTE): How Much Delay in Reality? • Comparable to radio latency L : 193.8ms-6.4s (29.8-69.6%) 2,000 2,000 2,000 2,000 Total suspension Total suspension Total suspension Suspension (ms) Suspension (ms) Suspension (ms) Suspension (ms) 1,500 1,500 1,500 1,500 Routing update Routing update State transfer (ctrl → ctrl) State transfer (ctrl → ctrl) 1,000 1,000 1,000 1,000 Radio conn. setup 500 500 500 500 0 0 0 0 0 0 0 0 25 25 25 25 50 50 50 50 75 75 75 75 100 100 100 100 Normalized sorted sample (%) Normalized sorted sample (%) Normalized sorted sample (%) Normalized sorted sample (%) 19
Option 1 (4G LTE): How Much Delay in Reality? • Comparable to radio latency L : 193.8ms-6.4s (29.8-69.6%) • Failure recovery incurs extra delay L : 500ms-30s 2 2 2 T-Mobile AT&T Sprint Project-Fi 2,000 2,000 2,000 2,000 Normal data suspension (s) Normal data suspension (s) Normal data suspension (s) Total suspension Total suspension Total suspension Suspension (ms) Suspension (ms) Suspension (ms) Suspension (ms) 1.5 1.5 1.5 1,500 1,500 1,500 1,500 Routing update Routing update State transfer (ctrl → ctrl) State transfer (ctrl → ctrl) 1,000 1,000 1,000 1,000 Radio conn. setup 1 1 1 w w o o r r 500 500 500 500 s s e e 0.5 0.5 0.5 0 0 0 0 0 0 0 0 25 25 25 25 50 50 50 50 75 75 75 75 100 100 100 100 Normalized sorted sample (%) Normalized sorted sample (%) Normalized sorted sample (%) Normalized sorted sample (%) 0 0 0 0 0 0 0.5 0.5 0.5 1 1 1 1.5 1.5 1.5 2 2 2 20 Extra suspension for failure recovery (s) Extra suspension for failure recovery (s) Extra suspension for failure recovery (s)
Option 2: Availability Over Consistency • Delivery user data without latest data session states • No guaranteed correctness : sequential consistency violated ? ? Internet Data-plane Control-plane Session state transfer ? ? Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 21 Gateway (eNB) (MME) (HSS) (SGW/PGW)
Summary: Impact of CAP Theorem Consistency over Availability Availability over Consistency • Guaranteed correctness J • Correctness unguaranteed • Long data service delay L • Immediate data service J 22
How to Balance Properties? Between availability, consistency and fault tolerance 23
Correctness, Not Sequential Consistency! • Sequential consistency is more than necessary for correctness Partition Availability Correctness Tolerance ≠ More flexible consistency model to achieve all! Sequential Consistency 24
What Consistency is Required for Correctness? • Function 1 : Data forwarding to correct location • Only a subset of network nodes need latest user location Internet Data-plane Control-plane Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 25 Gateway (eNB) (MME) (HSS) (SGW/PGW)
What Consistency is Required for Correctness? • Function 2 : Volume-based data billing • Decouple packet counting and billing policy • Packet counting can be performed together with data forwarding 1 + ? Internet Data-plane 1 0 Control-plane Tracking Area 1 Tracking Area 2 Base station Mobility controller User profile database 26 Gateway (eNB) (MME) (HSS) (SGW/PGW)
Other Cellular Functions in Paper • Radio access control • Cryptography over the air • QoS policy 27
Future Directions • What is the minimal consistency model for mobility support? • Can we have a continuous (thus tunable) balancing mechanism? • Will unreliable wireless connectivity affect the tradeoffs? 28
Conclusion • No free lunch for perfect mobility support • Sequential consistency is more than necessary for correctness • More flexible balance between C, A and P can benefits 5G 29
Backup 30
What Consistency is Required for Correctness? • Function 3 : Radio access control • Observation: per-group access control in 4G LTE • No need to wait for per-device radio access control list 4G Radio Access Control List Granularity Forbidden Tracking Area List (TS24.301) Per tracking area Forbidden PLMN list (TS24.301) Per roaming network Closed subscriber group (TS29.281) Per cell (private Femtocell) 31
Recommend
More recommend