Virtual Active Networks Gong Su Mar. 9, 2000

Network Computing Models Traditional: end-to-end, Client-server software at end nodes The network is but a packet-transport wire Emerging: edge-to-edge Application services/components deployed at edge nodes Examples: web proxies, firewalls, QoS/bandwidth brokers… Applications need to interact with network resources & topology Configure resources to provide appropriate service Adapt to availability and performance of network components

VAN: Middleware for Edge-Computing VAN is a middleware architecture that enables applications to Configure network topology Allocate node and link resources

A Driving Example A web caching application needs… Coverage for certain network area Connectivity among caching service components Resources to move cached contents Solution: requests a VAN that provides Coverage: spans a ring between AS1, AS2, AS4, and AS5 Resources: provides at least 1mbps for all connections Reliability: prohibits more than 2 virtual links from traversing the same physical link A DC B AS1 AS2 AS5 AS4 A AS3 B C D 1mbps Virtual spec. Physical network Mapping by VAN A DC B 1mbps Logical hierarchy

VAN Contributions Enable applications to configure network Algorithm that maps VAN specification to physical resources Acquire distributed node and link resources Deadlock-free VAN resource provisioning protocol Recover from underlying network failure Protocol that preserves VAN service semantics under failures

VAN Service Arch Components  VAN Local Manager (VLM)  Manages local node resources  Supports deadlock-free VAN provisioning  Monitors & reports resource status  VAN Domain Server (VDS)  Provides VAN services to application  VAN provisioning  Resource acquisition  Performance monitoring  Manages VAN to recover from physical network failure AS1 AS2 AS5 AS4 A AS3 B C D VDS VDS administrative domain Active node with VLM

Specification Mapping  Heuristic mapping algorithm A DC B AS1 AS2 AS5 AS4 A AS3 B C D 1mbps  Sort VNs and PNs by degree; map VN to PN by degree-order  Mark all PL without enough bandwidth for the VLs as infeasible  Each PL has a “mapped-onto” counter, initially  pick a VL and map it to a physical path with lowest maximum counter among all PLs traversed  After each VL is mapped, increment counter and subtract available bandwidth for each PL; mark a PL infeasible as appropriate 1  Repeat until all VLs are mapped

Resource Acquisition Protocol Acquires node and link resources Intra-domain: VDS – VLM Inter-domain: VDS – VLM and VDS – VDS Deadlock among competing VANs for shared resources can occur because One VAN is built in many domains distributedly Many VANs are built in many domains simultaneously Example VDS1 and VDS2 build VAN1 and VAN2 in domain A respectively VDS3 and VDS4 build VAN1 and VAN2 in domain B respectively VAN1 preempts VAN2 in domain A VAN2 preempts VAN1 in domain B VDS1 VDS2 VLM1 VDS3 VDS4 VLM2 VAN1 VAN2 A B

Deadlock Prevention Protocol How does the solution work Assign “weight” to VNs and VLs Each VDS computes a “Progress Index” (PI), indicating “how much” a VAN has been built PIs are globally synchronized and used as the priority for preemption when conflict happens 1 VDS’es request resource VLM detects conflict and initiates arbitration VDS broadcasts to all other VDS’es requesting global PI synchronization Other VDS’es ack sync request VDS’es ack arbitration request VLM notifies VDS’es with the arbitration decision

Failure Recovery When a physical link fails, the VLs it carries must be restored First try Local repair Find an alternative path with adequate resources between the two disconnected AS’es Fast, and preserve original topology But May violate reliability constraint Alternative path may not exist Example Physical link between 2 and 3 goes down Alternative path goes through 2, 1, 4, and Reliability violation Physical link Virtual link

Failure Recovery: Global Repair Local repair may violate reliability constraints or it may not be able to find an alternative path Global repair Computes substitute VL based on global topology and resource information Reconstruct topology when local repair cannot; guarantee reliability constraint But Computationally expensive Communication delay between root VDS and local VDS’es Example Substitute VL computed between 5 and 6, replacing the VL going through 2, 1, 4, and VDS1 VDS2 VDS3 VDS1 VDS2 VDS3 Physical link Virtual link

Schedule End of Spring 2000 Efficiently obtain global topology and resource information Summer 2000 (first half) Heuristics for virtual specification to physical network mapping with constraints Summer 2000 (second half) Algorithm for computing dynamic priority (PI) and analyze conflict resolution protocol Fall 2000 (first half) Efficient local repair mechanism (study MPLS fast rerouting, ATM self-healing, etc.) Fall 2000 (second half) Incremental global repair mechanism