Applying CORBA Fault Tolerant Mechanisms to Network Management. B. Natarajan, F. Kuhns, and C. O Ryan

Transcription

1 Applying CORBA Fault Tolerant Mechanisms to Network Management Aniruddha Gokhale Shalini Yajnik Bell Laboratories Lucent Technologies Douglas Schmidt B. Natarajan, F. Kuhns, and C. O Ryan Distributed Object Computing Group Washington University at St. Louis

2 Typical Telecommunications Applications 1 Common Requirements Network Management QoS Monitoring Network visualization and control High Availability % typical Achieved through monitoring and redundancy

3 Network Management - The Big Picture 2 QoS Monitoring/Policing for Service Classes E2 NOC E3 Software Agent Architecture AGENT MIB Key Bandwidth AN Autonomous Network E# Edge Router N# Inernal Router NOC Avg. Delay Error Rate Network Operations Center AN A E1 DiffServ Cloud E4 N2 N1 SPC Probe N3 Control Network Overlay Feedback Messages E5 B Embedded TAO SPC Hardware Configuration Events to NOC User Traffic User Traffic AN Problems QoS Monitoring Data collection, Control and Visualization Solution DOVE

4 Fault Tolerance 3 Problems 1. Developing highly reliable software 2. Emphasis on components and applications for telecom Forces 1. Standard COTS middleware 2. Rapid development cycles 3. Multi-platform support (heterogeneity) Solution ACE/TAO/DOORS integration

5 Integrated ACE/TAO/DOORS Platform 4! DOORS A CORBA service that supports fault-tolerant CORBA Objects TAO (The ACE! ORB) High-performance, real-time CORBA-compliant ORB ACE! Portable OO framework for communications software Implements core patterns

6 Example Application: Load Balancing Service 5 RANDOM ROUND ROBIN Service Overview LOAD BALANCING SERVER 2: IOR := resolve (strategy) 1:bind (strategy, IOR) CLIENT Improve overall application perfomance by balancing load e.g., supports both random and round-robin strategies 3:get_item () Fault Tolerance Issues 1:bind (strategy, IOR) SERVER SERVER SERVER Single point of failure Replication for high availability Failure detection/recovery

7 Example Application Architecture: DOVE 6 Distributed Object Visualization Environment (DOVE) 1. QoS performance monitoring 2. Fault tolerance 3. Real-time processing 4. Multi-platform

8 Application Management with DOVE 7 Multimedia Service Management 1. Integrated application management 2. DOVE agent Event propagation 3. DOVE Browser Visualize server performance Customizable visualization 4. Fault tolerance through DOORS

9 ReplicaManager Super WatchDog WatchDog DOVE Agent (Backup) Fault Tolerance using DOORS WatchDog DOVE Agent (Backup) ReplicaManager Replica configuration and management WatchDog and SuperWatchDog 8 WatchDog DOVE Agent (Primary) DOVE Vis. Comp WatchDog - Local failure detection polling, heartbeat SuperWatchDog - Host-level detection Failure notification Client side failure handling Transparent switchover from Primary to Backup

10 Applying DOORS to Load Balancing Service 9 Client WatchDog App. Manager 6 5 Primary Naming Service poll Load poll WatchDog 3 Balancer Replica Manager 8 Checkpoint Server Backup Load Balancer Usage 1. Application registered with RM 2. RM registers appl. with WatchDog 3. WD starts Appl. and polls 4. RM generates IOR and binds to Name Service 5. Client resolves IOR from Name Service 6. Client talks to Primary 7. Primary fails 8. Client transparently talks to Backup

11 Performance Results MEASURING CLIENT 3. resolve () 5. start_timer () 7. stop_timer () TAO NAMING SERVICE shutdown () 7.307/8.899/8.076 msec 6. callback () WATCH DOG 2. start () LOAD BALANCER LOAD BALANCER 2. start () WATCH DOG PRIMARY BACKUP 1.register_service () 1. register_service () DIFFERENCE BETWEEN OPERATIONS 7 & 5 M IN = m sec M AX = m sec A VG = msec REPLICA MANAGER

12 ' & '& & ' $ & % $ % $ % % Ongoing and Future Work Systematic empirical benchmarks to measure impact of fault tolerance on efficiency and predictability ' 11 $ Integrate DOORS/TAO with a variety of telecom applications Enhance DOORS and TAO to conform to forthcoming OMG CORBA fault tolerance spec (November 1999) Solicit feedback from Lucent Business Units

13 Overview Of CORBA Fault Tolerance Objectives Wide range of fault tolerance Simple low cost clients Highly reliable servers Passive and Active replication Automatic and application-controlled FT mechanisms Strong replica consistency Host1 Los Angeles FT Domain Host4 Wide Area Domain New York FT Domain C ORB Without FT Support IIOP Host2 A2 B1 Host3 Host5 B3 D1 Host6 D2 GW A1 A3 B2 Fault Tolerance RFP.html

14 Concluding Remarks 13 Summary Cost-effective reliability is important to next-generation telecom applications Benefits of DOORS/TAO integrated framework Provides the desired fault-tolerance Enables rapid application development Uses standards-based COTS CORBA middleware URLs