1 Praveen K. Muthuswamy Electrical Computer and Systems Engineering Rensselaer Polytechnic Institute...
-
Upload
caitlyn-towe -
Category
Documents
-
view
215 -
download
1
Transcript of 1 Praveen K. Muthuswamy Electrical Computer and Systems Engineering Rensselaer Polytechnic Institute...
1
ISPs as Nodes or Sets of Links?
Praveen K. MuthuswamyElectrical Computer and Systems Engineering
Rensselaer Polytechnic Institute
In collaboration withKoushik Kar, Aparna Gupta (RPI)
Hasan T. Karaoglu, Murat Yuksel (UNR)
2
Motivation
Contract-Switching Architecture
Internet Traffic Engineering using Contract-switching
Numerical study and Benchmarking
Distributed TE Solution
Contributions and Future work
Outline
Attain high-efficiency inter-domain TE
Numerous methods for intra-domain TE
Inter-domain TE involves many factors ranging from
technology to economics and policy
Current Inter-domain TE techniques are constrained to
outbound traffic load balancing
Cooperative inter-domain traffic engineering among
neighboring ISPs is necessary
Motivation
ISP is abstracted as a single node or as a set of nodes
(in case of multiple ASes)
Node abstraction simplifies inter-domain routing
Loss of path selection and flexibility beyond shortest-
path routing in terms of AS hops.
Sufficient flexibility at the routing level is crucial for TE
Internet Routing Limitations
Link-state mechanisms to inter-domain routing
◦ K. Levchenko, G. M. Voelker, R. Paturi, and S. Savage. Xl: an efficient network
routing algorithm. In Proc. SIGCOMM, 2008.
Pathlet Routing abstracts ISP as virtual nodes
◦ Scott Shenker, P. Brighten Godfrey, Igor Ganichev, and Ion Stoica, Pathlet routing.
Proc. SIGCOMM, 2009.
Inter-AS source routing and GMPLS
◦ Xiaowei Yang, David Clark, and Arthur Berger, NIRA: A new inter-domain routing
architecture. IEEE/ACM Transactions on Networking, 15:775–788, 2007.
◦ E. Mannie. Generalized mpls architecture, RFC 3945
Related Work
6
Contract-Switching Paradigm for Internet value flows and risk management, Yuksel et al. 2008
ISP is abstracted as a set of contract links
Contract link – Advertisable contract between edge nodes (peering points) of an ISP
Contract-Switching Paradigm
Network Coreaccessed onlyby contracts
Customers
EdgeRouter
EdgeRouter
EdgeRouter
EdgeRouter
EdgeRouter
EdgeRouter
Stations of the provider computing and advertising local prices for edge-to-
edge contracts.
30Mbps, 45 min, $9
7
Each contract link follows a single, fixed intra-AS path End-to-end flows are optimally split along contract links Optimal inter-domain traffic engineering, but simple intra-
domain routing
Internet Traffic Engineering using CSP
Notations
- Bandwidth capacity of router-to-router link
- Physical links used to construct edge-to-edge link
- Total traffic at edge router for destination
- Flow on the contract or peering link
9
Maximum Throughput Routing
Optimal TE formulation (1/2)
Capacity constraints
Flow-conservation constraints
Edge-to-edge flows
Throughput
10
Minimum Delay Routing
Vxy(f) is a convex-cost with each router-to-router link. Consider M/M/1 delay cost.
Minimum Bandwidth Routing
Optimal TE formulation (2/2)
11
Global Optimum (OPT) – Complete Internet topology
without abstractions (optimal inter and intra-domain
routing)
BGP framework – Least ISP hops for inter-domain and
OSPF/RIP for intra-domain
Frameworks for Benchmarking
12
Random topology◦ Inter-domain and Intra-domain are random
BRITE topology◦ BRITE model for inter-domain◦ Rocketfuel Topologies (ABILENE and GEANT) for
intra-domain GTITM topology
◦ GTITM model for inter-domain◦ Rocketfuel Topologies (ABILENE and GEANT) for
intra-domain
Simulation Results
13
Average throughput on 50 random topologies
Maximum Throughput
14
Delay on Sample Random topology and Sample BRITE topology
Minimum Delay
Average total bandwidth on 50 BRITE topologies
Minimum Bandwidth
Derived using the gradient descent approach to penalized objective function
Traffic rate variables for each destination at each edge-link
Updating of traffic rates requires only “local” information Gradient projection method for AS j involves projection
on , the intra-domain capacity region Gradient projection method for e-edge links requires the
congestion along the link and local communication between the end nodes
Distributed TE solution
17
Generate Internet topologies according to BRITE Size of each AS is 10 routers Obtain ratio of converged cost to optimum cost
Both average and maximum ratio is close to 1 Converged cost of the proposed TE solution is close to
the optimum
Numerical Study
Number of ASes
10 20 30
Average 1.0039 1.0057 1.0059
Maximum 1.0046 1.01 1.0093
18
Average convergence time
Convergence if cost does not vary beyond 0.5% Convergence time seems to increase sub-linearly with
number of ASes, and number of destinations
19
Contract link abstraction provides significant
improvement in routing performance over BGP
Close to the best achievable performance
Link abstraction is good for inter-domain traffic
engineering
Developed distributed TE solution based on gradient
descent and studied its performance numerically
Conclusion
Thank You !!Questions ?