Source-linked AI summary
Follow-Me Cloud: When Cloud Services Follow Mobile Users
T. Taleb, A. Ksentini, P. Frangoudis
TL;DR
Mobile users may retain optimal access to data-anchor gateways while receiving services from distant, suboptimal data centers. The paper proposes Follow-Me Cloud, combining mobility-aware service migration with LISP- or SDN-based continuity mechanisms; analytic and testbed results demonstrate its feasibility and advantages, while implementation choices retain concrete continuity limitations.
Problem
Optimal mobile data-anchor connectivity does not ensure optimal end-to-end service delivery when users move and continue receiving services from distant data centers.
Method
Follow-Me Cloud combines an MDP-based service-migration algorithm with alternative LISP- and SDN-based schemes for service continuity across decentralized cloud infrastructure.
Results
Analytic evaluation and testbed experiments demonstrate the feasibility and advantages of FMC and show that delaying migration lowers optimal-DC connectivity while increasing distance and latency.
Takeaways & Limitations
FMC provides a framework for keeping mobile users connected through suitable gateways while delivering services from cloud locations selected to support user experience and resource-aware operation.
Takeaways & Limitations
One continuity approach can break current VM connections, while an alternative requires modifying the hypervisor and uses additional LISP signaling.
Abstract
from arXiv · showhide
The trend towards the cloudification of the 3GPP LTE mobile network architecture and the emergence of federated cloud infrastructures call for alternative service delivery strategies for improved user experience and efficient resource utilization. We propose Follow-Me Cloud (FMC), a design tailored to this environment, but with a broader applicability, which allows mobile users to always be connected via the optimal data anchor and mobility gateways, while cloud-based services follow them and are delivered via the optimal service point inside the cloud infrastructure. FMC applies a Markov-Decision-Process-based algorithm for cost-effective, performance-optimized service migration decisions, while two alternative schemes to ensure service continuity and disruption-free operation are proposed, based on either Software Defined Networking technologies or the Locator/Identifier Separation Protocol. Numerical results from our analytic model for FMC, as well as testbed experiments with the two alternative FMC implementations we have developed, demonstrate quantitatively and qualitatively the advantages it can bring about.
1 INTRODUCTION
Mobile traffic growth and federated clouds motivate decentralized, network-aware service delivery, but user movement can leave services anchored at distant data centers. Follow-Me Cloud addresses this gap by migrating services toward optimal cloud locations while balancing user experience and migration cost.
- Explosive mobile data traffic is driving operators toward decentralized user-plane architectures with edge data anchors and traffic offloading.
- Federated clouds distribute virtual resources across geographically dispersed, potentially heterogeneous data centers to improve proximity and coverage for mobile users.
- The central challenge is that optimal connectivity to a mobile data anchor does not guarantee optimal end-to-end service delivery when users move away from the serving data center.
- Follow-Me Cloud lets services follow mobile users, maintaining optimal data-anchor and mobility gateways while selecting service points using proximity, load, delay, or other metrics.
- The framework combines service-continuity mechanisms, an MDP-based migration algorithm, analytic evaluation, and testbed implementations based on LISP and SDN.
2 RELATED WORK
Related work addresses service and user mobility through identifier-location separation, hypervisor integration, VM migration, and SDN control. FMC builds on these directions while combining LISP- and SDN-based continuity mechanisms with broader migration decision support.
- OpenFlow-based continuity methods can face scalability challenges involving flows, rules, setup rates, and control-channel bandwidth.
- Existing mobility approaches integrate IP mobility into hypervisors or invoke Mobile IP during VM lifecycle events, but live migration can incur potentially long downtimes.
- Modified LISP can achieve sub-second VM migration downtimes, although one cited scheme requires hypervisor modification and control-plane signaling.
- LISP separates endpoint identifiers from routing locators, allowing a station to move without changing its endpoint identifier while updating its routing locator.
- FMC extends earlier work with a generic decentralized-mobile-network design, heavier NFV use, an updated LISP implementation, an SDN implementation, and an extended migration algorithm.
3 THE FOLLOW-ME CLOUD CONCEPT
FMC is a modular architecture for delivering cloud services from locations close to mobile users in decentralized mobile and federated-cloud environments. Its migration process selects when and where to move services according to mobility, service, and network considerations.
- 3.1 High-level design: FMC consists primarily of an FMC controller and a data-center/gateway mapping entity, which may be independent, collocated, or software components running on a cloud data center.
- 3.1 High-level design: The design is generic beyond LTE/EPS and maps data centers to data-anchor routers such as PGWs or Wi-Fi access routers.
- 3.1 High-level design: Services can be brought to data-anchor routers through micro-data centers, while macro-data centers provide persistent VM storage and service support.
- 3.1 High-level design: For simplicity, the study assumes services start at the user’s attached data anchor and migrate without replication as the user moves.
- 3.2 Service migration process: Migration may be triggered by data-anchor changes and depends on service requirements, content size, user class, QoE, cost, load, resource utilization, and micro-data-center capacity.
- 3.2 Service migration process: The FMC controller selects the destination micro-data center and determines whether a multi-component service should be fully or partially migrated while accounting for migration cost.
4 AN ANALYTIC MODEL FOR FMC
The analytic model represents user mobility and service migration in a Markov framework, then quantifies distance, delay, migration cost, and service disruption. State aggregation reduces model complexity while retaining the behavior needed to evaluate FMC tradeoffs.
- 4.1 Markov-based system model: The model uses a random-walk UE mobility process and a CTMC whose states encode the UE’s distance and cell position relative to the serving DC.The UE chooses each of six neighboring cells with probability p = 1/6, remains in a cell for an exponentially distributed time with mean 1/µ, and triggers migration at k hops.
- 4.1 Markov-based system model: State aggregation reduces the CTMC state-space explosion by grouping states with identical transition behavior into a lower-state Markov chain.For k = 5, ring-specific transition patterns determine the aggregated states and their steady-state probabilities.
- 4.2 Average UE-DC distance and the probability to be connected to the optimal DC: The model derives average UE-to-serving-DC distance and the probability π0 that the UE is connected to the optimal DC during the system lifetime.The UE remains connected to the current data anchor until service migration is triggered.
- 4.3 Average end-to-end delay from the serving DC: Average end-to-end delay E[D] measures the time for a UE to receive packets from the serving DC and depends on the UE’s distance from its data anchor router.Di denotes the end-to-end delay when the UE is in ring i.
- 4.4 Service migration cost: Migration cost includes transferred object size and signaling overhead, while service disruption duration depends on transfer size, DC-to-DC RTT, packet loss, and VM conversion.FMC uses three signaling messages to trigger migration, and the disruption time is modeled with an empirical TCP latency model.
5 A MDP-BASED SCHEME FOR SERVICE MIGRATION
The service-migration policy is formulated as a CTMDP/DTMDP that balances migration cost against user-perceived quality. States represent distance from the current data center, and actions determine whether service migration occurs.
- Actions: Action a2 migrates the service to an optimal data center, whereas action a1 keeps the user served by the same data center.
- 1D mobility model: In the 1D model, states range from 0 to thr, where thr is the maximum visited-SA distance before migration to the optimal data center.
- The CTMDP associates each state with available actions, transition probabilities, and rewards for service-migration decisions.
- Solution method: Uniformization converts the continuous-time process into a discrete-time model, enabling discounted optimality equations and value-iteration solution for the optimal policy.
- Migration threshold: When the user enters a service area beyond threshold thr, migration is automatically triggered because only action a1 is available at state thr.
- Reward model: The reward combines user quality, which decreases with hop distance, and the cost Cm of migrating all or part of the service.
6 IMPLEMENTATION ALTERNATIVES
FMC is implemented through alternative SDN- and LISP-based architectures that detect mobility, select service locations, and support service migration across distributed data centers. The implementations use controller-driven routing or locator updates to maintain service delivery as users move.
- The paper explores SDN and LISP as alternative technologies for realizing FMC, while noting that Mobile IPv6 is also compatible with the design.
- SDN-based implementation: In the SDN architecture, a NOX-based FMCC coordinates OpenFlow access routers, micro-data-center routers, virtual switches, and cloud hypervisors.
- Location management: The SDN FMCC selects a micro-data-center using geographic location and client-to-VM path characteristics such as delay and congestion.
- Mobility and migration: After mobility detection, the SDN controller executes the migration decision algorithm, installs forwarding rules when needed, and launches the migration process.
- Session management: The SDN controller preserves ongoing sessions during VM migration by avoiding VM configuration changes and establishing a tunnel within the visited network segment.
- LISP-based implementation: In the LISP architecture, mobility updates pass from the new subnet’s xTR through the MR/MS to the FMCC, which decides whether to migrate the service to the corresponding micro-data center.
- Service continuity: The proposed LISP continuity approach informs both involved xTR routers about the VM’s changed RLOC after migration.
7 PERFORMANCE EVALUATION
The evaluation combines analytic model results with SDN- and LISP-based testbed measurements, showing how FMC balances service quality, migration cost, and continuity. Delaying migration reduces cost but worsens proximity and latency, while implementation measurements identify short downtime and controller-scalability constraints.
- Model-based performance results: FMC evaluation measures optimal-DC connectivity, distance, connection latency, migration cost, and service disruption time using analytic and testbed-based results.The analytic model evaluates five performance dimensions, while experiments examine SDN- and LISP-based implementations.
- Model-based performance results: High migration thresholds reduce optimal-DC connectivity and increase average distance; the distance exceeds two hops when k is higher than 6.Migrating after each handover keeps the UE connected to the optimal DC, whereas delayed migration increases distance.
- Model-based performance results: About 200 ms separates the average connection delays for k = 2 and k = 7, with latency increasing as the migration threshold grows.The model uses Lati = 0.02i^2 (s), where i is hop distance from the serving DC.
- Model-based performance results: Migration cost decreases with k but is highest when all service content is migrated, because cost depends critically on the object size transferred.The evaluation compares migrating all, 50%, and 10% of the service.
- Service migration policies: For p = 0.8, the optimal policy migrates at d = 6 when τ = 0.1 and d = 5 when τ = 0.5; higher τ produces fewer migrations.For a random walk, p = 0.5, both scenarios trigger migration when distance exceeds 5.
- Testbed-based performance results: OpenFlow rule installation increases initial delay and implies a scalability issue for large centralized FMC deployments.The paper identifies decentralization as a possible approach to address this issue.
- Testbed-based performance results: 7.5 ms of service downtime occurred during a 26-second SDN-based migration, while the client latency converged near 1 ms after migration.The experiment migrated service from DC1 to DC2 after the measured RTT increased to approximately 250 ms.
- Testbed-based performance results: LISP-based downtime increases with controller-to-target latency, whereas VM migration duration becomes practically independent of DC-to-DC RTT beyond 10 ms.The downtime is tied mainly to mobility-management notification, while migration time is more affected by the 100 Mbps link bandwidth.
8 CONCLUSION
Follow-Me Cloud enables cloud services to follow mobile users through migration decisions that balance cost and user experience. Analytic results and SDN- and LISP-based testbeds demonstrate optimized delivery and deployment feasibility.
- Follow-Me Cloud migrates service resources to appropriate cloud locations as users move.The framework targets enhanced cloud-service delivery for mobile users and optimized user experience.
- Its Markov-Decision-Process-based algorithm balances migration cost against user experience when deciding whether to migrate services.
- Two alternative architecture designs support FMC service continuity: one uses SDN technologies and the other uses the LISP protocol.
- Numerical model results and SDN- and LISP-based testbed experiments demonstrate optimized mobile cloud-service delivery and feasibility for real-world deployment.