Multipath Routing
CS 522
F2003
Beaux Sharifi
Agenda
•
•
•
•
•
Description of Multipath Routing
Necessity of Multipath Routing
3 Major Components Necessary for Multipath Routing
Example Multipath Routing Model
Simulation Results
What is Multipath Routing?
• Multipath Routing is the spreading of traffic from a source node to a
destination node over multiple paths through the network.
Figure 1. Multipath Routing Model Diagram.
Importance of Multipath Routing
• The Internet is a significant part of the Global communication
infrastructure.
• The use of the Internet is growing at an incredible rate:
– Jan 1999: ~ 43 million hosts
– Jan 2003: ~ 171 million hosts (source: Internet Domain Survey)
• QoS, throughput, and delay are difficult problems with current singlepath routing architecture.
• From queuing theory, we know that through increased sharing, overall
utilization of the entire network is improved.
• Multipath routing provides much better overall network performance
by allowing better sharing of the available network resources.
3 Major Components
• A Multipath Calculation algorithm to compute
multiple paths.
• A Multipath Forwarding algorithm to insure that
packets travel on their specified paths.
• An End-Host Protocol that effectively uses the
determined multiple paths.
Path Algorithms
• Generate paths based on a the desired characteristics of the path.
– i.e. Maximized throughput or minimized delay
• Generate Multi-Option paths and/or Multi-Service paths.
• Path requirements depend on the end-user application.
– i.e. Telnet vs. FTP
• Two characteristics of a quality path:
– Path Quantity
– Path Independence
Path Algorithms (cont.)
• Some path algorithms that don’t work:
– Shortest K Paths, Link Disjoint Paths, Maximum Flow
• Two path algorithms that do work:
– Maximize Throughput: Capacity Removal
– Minimize Latency: Discount Shortest Path
• Both algorithms based on Dijkstra’s Shortest Path algorithm.
• Both algorithms produce shortest paths with minimal overlap by
incrementally adding “cost” to each of the previously found paths.
Path Forwarding
• Path Forwarding Problem: how to specify a packets path and then
forward packets along that path.
• Each router has potentially multiple routes to a destination node.
• The destination address is no longer sufficient.
• A Path Identifier is now required for every packet.
• Design Requirements for Path Forwarding:
– Minimize Packet Overhead
– Minimize router CPU overhead of forwarding packets
– Minimize additional router memory
End-Host Protocol
• Performance gains are only realized if end-hosts use the multiple paths
effectively.
• Paths can be used concurrently or one at a time.
• The appropriate use of multiple paths is application specific.
– Instant Messenger (multi-service)
– Urgent Message (multi-option)
Example Multipath Routing Model
• Developed by Johnny Chen of Rice University in 1999.
• Consists of two different routing algorithms based on extensions of the
traditional routing algorithms:
– MPDV (MultiPath Distance Vector)
– MPLS (MultiPath Link State)
• Both routing algorithms seek to optimize throughput by using a
Capacity Removal based algorithm.
• Chen develops efficient path forwarding algorithms while minimizing
packet and router overhead.
• Uses a fixed-length packet path ID to provide minimal packet
overhead and allow efficient indexing into router forwarding tables.
Example Multipath Routing Model
(cont.)
• Contains a new transport layer called MPTCP (Multipath TCP).
• MPTCP is based on single-path TCP and provides a reliable bit stream
service.
• MPTCP operates by opening multiple TCP connections on different
paths and then multiplexing data between them.
• The receiving MPTCP layer collects data from each of the connections
and then restores the original message stream.
• MPTCP provides flow and congestion control.
• MPTCP provides increased network performance without any changes
to user-applications.
Simulation Results
• Chen compares both MPDV and MPLS using a packet-level network
simulator “xsim” from the University of Arizona.
• Simulated network is similar to the Internet topology with 100 nodes
and 195 links across multiple clusters.
• Performance is measured in throughput, latency, and message drop-off
probability.
• Throughput is measured using MPTCP.
• Latency and drop-off probability is measured using multipath ping.
Throughput Results
Figure 2. Foreground MPTCP Performance using MPDV and MPLS.
Latency and Message-Drop Results
Figure 3. Latency and Message Drop Percentages with MP-ping and SP-ping.
Summary
• MPDV and MPLS demonstrate that multipath routing provides higher
performance than their single-path counterparts.
• Performance results are consistent across different network topologies
and network utilization levels.
• The cost incurred by implementing a multipath routing algorithm
doesn’t outweigh the added performance benefits.
References
• Chen, Johnny. New Approaches to Routing for Large-Scale Data
Networks. Rice University, Houston, Texas: PhD Thesis, June
1999.
• Nua Internet How Many Online. Nua.com 11 Dec. 2003
<http://www.nua.ie/surveys/how_many_online/>
• Tanenbaum, Andrew S. Computer Networks, Fourth Edition.
Upper Saddle River, New Jersey: Prentice Hall PTR, 2003.
Questions?