1 / 17

Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks*

Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks*. Dr. Spiros Bakiras, PhD (USC), MIEEE Dept. of Electrical & Electronic Engineering The University of Hong Kong *Joint work with Prof. Victor O.K. Li. Outline. Motivation System design Resource management

bena
Download Presentation

Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks*

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks* Dr. Spiros Bakiras, PhD (USC), MIEEE Dept. of Electrical & Electronic Engineering The University of Hong Kong *Joint work with Prof. Victor O.K. Li

  2. Outline • Motivation • System design • Resource management • Simulation results • Conclusions Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  3. Motivation • The Differentiated Services (DiffServ) architecture has been proposed by the IETF as a scalable solution for providing end-to-end QoS guarantees. • Scalability at the data plane • Edge routers manage traffic on a per flow basis. • Core routers manage traffic on a much coarser granularity. • Only a limited number of service classes is defined, and each packet is forwarded based on Per Hop Behavior (PHB). • Examples: Expedited Forwarding (EF), and Assured Forwarding (AF) PHB. • Scalable resource management is still an open research problem. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  4. Motivation • Initial proposal: centralized agent, called Bandwidth Broker (BB). Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  5. Motivation • Disadvantages • The signaling messages will create congestion on the links around the BB. • The BB must maintain per flow information about every flow which is active inside the domain. • The BB is a single point of failure. • We will propose a decentralized resource management scheme, where • The BB is eliminated from the admission control procedure. • All the edge routers in a domain will independently set-up and tear-down individual connections. • The BB will periodically reallocate the resources inside the domain. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  6. System design • Intra-domain routing • Static routing tables for QoS-sensitive traffic. • A total of k paths are pre-computed for each pair of edge routers. • Each QoS flow will be assigned to one of these paths. • Inter-domain routing • All the edge routers in the domain will use the standard BGP protocol to exchange reachability information with neighbor domains. • Packet forwarding with IPv6 • The flow label (20 bits) will indicate the corresponding ingress-egress pair and path number. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  7. Resource management • Define • SD pair: a pair of ingress-egress routers. • For each SD pair there are k pre-computed paths. • The source of each SD pair i will periodically measure the total amount of reserved bandwidth r[i]. • Every time interval T, the BB will receive the following information about every SD pair i • The current reserved bandwidth x[i][j] for each path j. • The average reserved bandwidth w[i] over the measurement window. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  8. Resource management • The BB will then calculate the new amount of resources that should be allocated to each path. • Each path jof SD pair i will be allocated an amount of bandwidth y[i][j], where y[i][j]x[i][j]. • Rejected traffic load • Instead of the actual reserved bandwidth r[i], we will periodically measure the virtual reserved bandwidth v[i]. • This measurement will include the bandwidth of all the rejected flows. • w[i] will then represent the average virtual reserved bandwidth over the measurement window T. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  9. Resource management • Definitions • Every SD pair i will require an additional amount of bandwidth b[i]=w[i]-r[i]. • Eligible SD pair: a pair i with b[i]>0. • Saturated SD pair: a pair for which there is at least one saturated link in each of the k paths. • Resource reallocation algorithm • In a round-robin manner, add a small amount of bandwidth dx to SD pair i, to the path j that minimizes a cost function. • Repeat until all SD pairs become either non-eligible or saturated. • Allocate the unused bandwidth to all unsaturated SD pairs. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  10. Resource management • Admission control • The source will send a PATH message to the destination. • Admission control will be performed at the ingress router of each domain. • If the new flow is admitted, the destination will reply with a RESV message. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  11. Resource management • Advantages • The signaling overhead is spread across multiple links. • Per flow information is distributed across the edge routers of the domain. • A failure of the BB or an edge router will not affect the functionality of the domain. • Fast response time for new connection requests. • Equal to the round-trip delay between the source and the destination. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  12. Simulation results • We simulated our resource management scheme in the MCI internet topology. • It consists of 19 nodes and 32 links. • We assumed all links have a capacity of 2.4 Gbps. • New requests arrived at each SD pair, according to a Poisson process with average rate that was uniformly distributed in [1,rmax]. • Every new request was either for a voice or video connection with equal probability. • Voice: bandwidth 64 kbps, duration 5 min (exponential). • Video: bandwidth uniform in [0.3,2] Mbps, duration 20 min. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  13. Simulation results Fig. 1: The simulated network topology Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  14. Simulation results Fig. 2: Bandwidth blocking rate as a function of the average arrival rate, for different values of k. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  15. Experimental results Experimental results Fig. 3: Bandwidth blocking rate as a function of the average arrival rate, for different values of T. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  16. Simulation results Fig. 4: Bandwidth blocking rate as a function of the average arrival rate, for different resource management schemes. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

  17. Conclusions • We have proposed a new resource management scheme for end-to-end QoS guarantees in DiffServ-based networks. • All the edge routers of a domain actively participate in this procedure. • The BB is only responsible for periodically reallocating the resources inside the domain. • Admission control is performed instantaneously at the ingress router of each domain. • The simulation results indicate that our scheme can achieve lower bandwidth blocking rate compared to traditional shortest path routing. Efficient Resource Management for End-to-End QoS Guarantees in DiffServ Networks

More Related