1. APNOMS2003Fujitsu Laboratories Ltd.1 A QoS Control Method Cooperating with a Dynamic Load Balancing Mechanism Akiko Okamura, Koji Nakamichi, Hitoshi Yamada and Akira Chugo Fujitsu Laboratories Ltd. 4-1-1, Kamikodanaka, Nakahara, Kawasaki, 211-8588, Japan Telephone: +81-44-754-2635 Fax: +81-44-754-2741 E-mail address: {akikoo, nakamichi, hitoshi, chugo}@flab.fujitsu.co.jp

2. APNOMS2003Fujitsu Laboratories Ltd.2 Introduction Current IP network problems –End user's viewpoint Degraded of performance (e.g., lowered throughput and increased delay) due to congestion –Network operator's viewpoint Profits do not improve though traffic increases every year Frequent bandwidth increases needed to support traffic increases Services supporting usage-based billing are limited To overcome these problems –Provide QoS guaranteed service –Develop fee/charge system based on network QoS –Use network resources efficiently --> IP traffic control mechanism is required

3. APNOMS2003Fujitsu Laboratories Ltd.3 Our Approach to IP Traffic Control QoS: quality of service SLA: service level agreement SPF: shortest path first Level of network efficiency Traffic Engineering (TE) Minimum hop routing +Explicit routing +Multi route & path control +Dynamic flow splitting Best Effort Packet- priority mechanism Admission policing shaping Performance monitoring reporting Account- refund Connectionless approach (Diffserv) Connection- oriented approach (IntServ) Explicit path setup Static load balancing Dynamic load balancing LowHigh Low High Level of QoS/SLA Future Hop-by-hop forwarding Proposed method

4. APNOMS2003Fujitsu Laboratories Ltd.4 Traffic Engineering (TE) Description –Improves traffic performance –Facilitates reliable network operations –A main application of multiprotocol label switching (MPLS) Constraint-based routing (explicit routing) Example Applications –Static/dynamic load balancing Achieves highly reliabile network by avoiding congestion/failure Enables efficient use of bandwidth resources --> Functions of dynamic load balancing have been proposed –Fast Reroute Achieves highly reliabile network through high-speed failure recovery

5. APNOMS2003Fujitsu Laboratories Ltd.5 Basic Architecture of Proposed Method - Cooperation between dynamic load balancing and providing QoS guarantee - TE Controller QoS path control for guaranteed class traffic QoS path Bandwidth reserved High priority at scheduling Optimum route considering both network and application server resources MPLS: multi-protocol label switching MPLS Network Best Effort path Original path = minimum hop route Detour route used when there is congestion Application Server Statistics monitoring (Network, Application servers) User QoS request Load balancing control for best effort traffic Admission control

6. APNOMS2003Fujitsu Laboratories Ltd.6 QoS Routing Algorithm Minimize total cost of link and server total cost = server cost + link cost server cost: 1/(residual available output rate) link cost: 1/(residual available bandwidth) link cost server cost User TOTAL COST is MINIMUM! Link D-E cost is high… A B C D E server cost is high…

7. APNOMS2003Fujitsu Laboratories Ltd.7 Evaluation of QoS Routing:Metrics and Model Metrics –Number of QoS requests accepted –Average number of hops in QoS paths Compared with LSL method (lowest server load) DNS method (domain name server) Simulation model –ISP network, 19 nodes –Application servers Four Capacity of 500 Mbps –QoS requests 1-10 Mbps bandwidth guarantee (random) Users edge selected at random (users edge servers edge) OC3 (155 Mbps) T3 (45 Mbps) Server location candidate Select server with lowest load Select minimum cost route to server Select nearest server 1 st step 2 nd step

8. APNOMS2003Fujitsu Laboratories Ltd.8 Evaluation of QoS Routing: Results Effect of QoS routing considering both server and network loads –Accommodates many more requests –Provides QoS path with the smallest number of hops 0 100 200 300 400 Number of QoS requests Number of requests accepted 0200400600 800 0200400600800 Number of QoS requests 2.8 3.0 3.2 2.0 2.6 2.4 2.2 1.8 Average number of hops DNS method Our Proposal LSL method

9. APNOMS2003Fujitsu Laboratories Ltd.9 Evaluation of Dynamic Load Balancing: Metric, Model, and Conditions Evaluate effect of dynamic load balancing under GS traffic conditions Metric Throughput of BE traffic Model 5-node-ring model with 100 Mbps links Conditions –50 Mbps BE traffic –Bandwidth reserved for GS traffic is increases to 80 Mbps. –Actual amount of GS traffic fluctuates –Congestion detection conditions A)Actual (GS+BE) traffic > 80 Mbps B)(Reserved GS + Actual BE) traffic > 80 Mbps 100 Mbps BE: 50 Mbps GS Reserved Actual GS traffic

10. APNOMS2003Fujitsu Laboratories Ltd.10 Evaluation of Dynamic Load Balancing: Results Time [s] Throughput [Mbps] Actual GS traffic BE traffic with D-LB D-LB1 (condition A, solid line) D-LB2 (condition B, × signs) BE traffic with D-LB D-LB1 (condition A, solid line) D-LB2 (condition B, × signs) BE traffic without D-LB Bandwidth reserved for GS traffic Throughput decreases because load cannot be moved to other available links. Input BE Traffic = 50 Mbps (fixed) Throughput remains almost maximum. D-LB:Dynamic load balancing

11. APNOMS2003Fujitsu Laboratories Ltd.11 Implementation Path-setting status display - Snapshot of Operation Screen - Click Control status display panel (Path setting, load balancing, etc.) Detailed path information

12. APNOMS2003Fujitsu Laboratories Ltd.12 Conclusion & Future Work Conclusion Proposed method effectively utilizes resource while providing QoS QoS routing based on network and server loads –Number of QoS requests accepted is improved –Server and network load balancing are achieved Use of dynamic load balancing effectively provides QoS-guaranteed service –Degradation in BE traffic throughput when GS traffic is fluctuating is avoided Future work Evaluation of performance in large-scale network Development of more advanced QoS control method based on TE