1. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Enhanced General Switch Management Protocol Salim Hariri Department of Electrical and Computer Engineering HPDC Laboratory The University of Arizona ECE Building, Room 421 Tucson, Arizona, AZ 85750 Tel: (520) 621-4378, Fax: (520) 621-8076 hariri@ece.arizona.edu, www.ece.arizona.edu/~hpdc

2. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Why Programmable Networks? l Rapid creation, deployment and management of new services in response to user demands. l Change in the nature of traffic due to the wide variety of applications and services. l Application specific demands for resources. l Need for the separation of communication hardware from control software. l Better control over the network resources for its effective use.

3. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Classification of Programmable Networks Programmable Networks IP ATM Active Networks (Dynamic Approach) q-GSMP GSMP e-GSMP P 1520 Model Discrete Approach (Out of Band) Integrated Approach (In band) Open Interface Networks (Static Approach)

4. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc l Provides abstractions in the layers of a node to define programmable interfaces. l Allows applications and middle-ware to manipulate low-level network resources. l Uses APIs to control the various layers. Open Interface Networks

5. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Open Interface Networks (Contd.) l Advantages: –Separation of service business. –Separation of vendor business. –Faster standardization. –Extensibility –Richer Semantics

6. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc l Quality of Service (QoS) refers to the capability of the network to provide better service to selected network traffic irrespective of the underlying technologies. l The goal of QoS is to provide priority including dedicated bandwidth, controlled jitter and latency and improved loss characteristics. Meaning of QoS

7. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc QoS in OSI Model

8. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc l Providing QoS over ATM is easier –Fixed length of cells –Well defined types of services. l Techniques for providing QoS in ATM –General Switch Management Protocol (GSMP) –q-GSMP QoS over ATM

9. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc General Switch Management Protocol(RFC 1987) l Open Interface, switch control protocol. l Connection oriented network technologies. l Point-point and Multi-point connections. l Adjacency protocol -synchronize state across link. l Master-Slave relationship between controller and the switches

10. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc GSMP ClassifierRegulatorScheduler VPC/VCCs Policer Output Port Confirming Excess QoS Class

11. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc GSMP Message Sequence Switch controller Network Switch Mapping Connection Management Port Management Configuration Information Statistics Events qGSMP ATM Switch

12. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc q-GSMP l Provides QoS extensions to the GSMP. l Supports new messages enabling selection of: – QoS constraints – Buffer management – Scheduling algorithms –Memory allocation schemes. l Specific to ATM switches.

13. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc q-GSMP Switch Controller Scheduler Network Switch Mapping Buffer Manager SR Estimator qGSMPATM Switch Connection Management Port Management Configuration Information Statistics Events QoS Management QoS Configuration Management QoS Statistics QoS Events

14. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc QoS over IP Issues: l IP was not meant to provide better than the best effort QoS. l Complex scheduling and buffer management due to variable length packets. l Changes in the traffic pattern at the output of a router due to traffic aggregation.

15. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc l Traditional IP networks support only best effort delivery. l Convergence of voice and data requires IP network to provide deterministic guarantees for real time traffic. l Multimedia traffic require both bandwidth and delay guarantees. Need for IP QoS

16. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc P-1520 Model l Defines a set of Programmable interfaces for the development of the protocol and management of the infrastructure. l Defines four interfaces: –Value Added Service Level –Network Generic Services Level –Virtual Network Device Level –Physical Elements Level

17. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc P-1520 Model Algorithms for value added communications Algorithms for routing and connection management Virtual Network Device Physical Elements (Hardware) V Interface U Interface L Interface CCM Interface Value Added Services Level Network Generic Services Level Virtual Network Devices Level PE Level

18. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc IP reference Model

19. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Features of our IP Model l Extension of Integrated Services Model of Clark,Shenker and Zhang[ ]. l Conforms to P1520 model[ ] of programmability. l Complies with ForCES architecture[ ] of network entity. l Incorporates ETRI’s Router Architecture.

20. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc ETRI’s Router Architecture

21. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Scheduler and Queues Outbound Packets Scheduler Packets in Various Queues

22. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Algorithm for Schedulability We consider the Diffserv QoS, classes of service. Let us assume that DS(i) = i th class of service. Vector Q(i) = [ q1(i) q2(i),…….qm(I)] be the QoS Parameters for the i th class. We define a matrix Qm consisting of all the QoS classes and their respective parameters. Q(1) q1(1) q2(1) ………qm(1) Qm = Q(2) = q1(2) q2(2) ………qm(2) Q(n) q1(n) q2(n) ………qm(n)

23. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Packet Classification Switch/ Backplane Media I/F 01 02 n-1 n Input of a Router Bandwidth/Delay Shaped Queues Per-dest Queues Scheduler Inbound Packets

24. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc A table K is maintained by the router to indicate the number of packets in each QoS class. K = [k1,k2…………kn] Bo = port bandwidth B = total bandwidth. Bo = Σ ki * q1(i) if (q1(i) <= (B-Bo)) then BandSchedulable =TRUE; else BandSchedulable = FALSE;

25. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Packet Processor Switch/ Backplane Output from a Router Bandwidth/Delay Shaped Queues Scheduler Media I/F Outbound Packets

26. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Admission Control Algorithm The Admission control is made at the line cards. Token Bucket algorithm is used to characterize the flows. A new flow is admitted if the following condition is satisfied PBR new + N Σ i=1 PBR i <= ρC where PBR = peak bit rate ρ = admissible load of capacity C = total capacity

27. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc e-GSMP l Envisages to provide QoS services over a primarily IP network. l Is an Open Interface approach. l Defines a Master-Slave relationship between the Controller and the IP Routers. l Allows an interactive approach to provide programmability.

28. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc e-GSMP Switch Fabric Input PortsOutput Ports Multiplexer 1 m Switch Mapping Schedulable Region Estimators

29. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc l To support the Intserv and Diffserv frameworks. l To define schedulable regions to implement admission control. l To allow for different scheduling and buffer management techniques. e-GSMP

30. Salim Hariri HPDC Laboratory http://www.ece.arizona.edu/~hpdc Concluding Remarks