New Routing ArchitecturesNew Routing Architectures

Jennifer RexfordJennifer Rexford

Advanced Computer NetworksAdvanced Computer Networkshttp://www.cs.princeton.edu/courses/archive/fall08/http://www.cs.princeton.edu/courses/archive/fall08/

cos561/cos561/Tuesdays/Thursdays 1:30pm-2:50pmTuesdays/Thursdays 1:30pm-2:50pm

Outline

• Changing the routing architecture– Why?– Where and how?

• Example architectures– Removing routing from routers– Hybrid Link-state/Path-vector– Resilient Overlay Routing

Why Change Routing?

• Better performance– Scalability, security, convergence, reliability,

flexibility, …

• Simpler management– For network operators– For folks deploying services

• Greater extensibility– To enable experimentation– To enable new services

What to Change, and Where?

• Add another layer about network routing– Routing functionality in overlay networks

• Change the routing protocols– To improve scalability, security,

convergence, …

• Change the division of functionality– Data, control, and management planes

• Change the division of responsibility– End users, third parties, and service

providers

• ???

Removing Routing from Routers: Routing Control Platform, Routing as a Service, 4D Control Plane, Ethane,

…

Network Operators

• Network-wide views– Network topology (e.g., routers, links)– Mapping to lower-level equipment– Traffic matrix

• Network-level objectives– Load balancing– Survivability– Reachability– Security

• Direct control– Explicit configuration of data-plane

mechanisms

What Should Routers Do?

• Forward packets: yes– Must be done at high speed– … in line-card hardware on fast routers– So, needs to be done on the routers

• Collect measurement data: yes– Traffic statistics– Topology information

• Compute routes: no???– Distributed computation of forwarding tables– Doesn’t inherently need to run on the routers

Reasons to Remove Routing From Routers

• Routing is hard to do in a distributed fashion– Beyond single-path and/or shortest-path routing

• Difficult to make load-sensitive routing stable– Over-reacting to out-of-date information

• Poor visibility to drive good decisions– Incomplete local views of topology and load

• Not flexible enough for end users– Cannot easily select customized routes

• Difficult to extend over time– Hard-coded into the underlying routers

Routing Control Platform

• Goal: Move beyond today’s artifacts, while remaining compatible with the legacy routers

• RCP computes routes for the routers– Network-wide visibility and control

• Backwards compatibility– RCP speaks to routers using BGP protocol

AS 2

RCP

Example Services

• Selective denial-of-service attack blackholing– Identify entry point and victim of attack– Drop offending traffic at the entry point

• Planned maintenance dryout– Drain traffic off of an edge router– Before bringing it down for maintenance

• Flexible egress point selection– Multiple ways to reach the same destination– Giving customers control over the decision

• Enhanced interdomain routing security– Anomaly detection or security protocols

Routing As a Service

• Goal: third parties pick end-to-end paths for clients to satisfy diverse user objectives

• Forwarding infrastructure– Basic routing (e.g., default routing)– Primitives for inserting routes

• Route selector– Aggregates network information– Selects routes on behalf of clients– Competes with other selectors for customers

• End host– Queries route selector to set up paths

Feasibility

• Fast reaction to failures– Routers are closer to the failures– Can a service react quickly enough?

• Scalability with network size– State and computation grow with the topology– Can a service manage a large network?

• Reliability?– Service is now a point of failure– Is simple replication enough?

• Security?– Service is now a natural point of attack– Easier (or harder) to protect than the routers?

Improving BGP Convergence

Routing Change: Before and After

0

1 2

3

0

1 2

3

(1,0) (2,0)

(3,1,0)

(2,0)

(1,2,0)

(3,2,0)

Routing Change: Path Exploration

• AS 1– Delete the route (1,0)– Switch to next route

(1,2,0)– Send route (1,2,0) to AS

3

• AS 3– Sees (1,2,0) replace

(1,0)– Compares to route (2,0)– Switches to using AS 2

0

1 2

3

(2,0)

(1,2,0)

(3,2,0)

Routing Change: Path Exploration

• Initial situation– Destination 0 is alive– All ASes use direct path

• When destination dies– All ASes lose direct path– All switch to longer

paths– Eventually withdrawn

• E.g., AS 2– (2,0) (2,1,0) – (2,1,0) (2,3,0) – (2,3,0) (2,1,3,0)– (2,1,3,0) null

1 2

3

0

(1,0)(1,2,0)(1,3,0)

(2,0)(2,1,0)(2,3,0)(2,1,3,0)

(3,0)(3,1,0)(3,2,0)

Convergence Overhead and Delay

• Path exploration is expensive– Large number of possible paths– Might have to explore (nearly) all of them

• Much slower than link-state routing– Simply floods the topology– And routers compute shortest path

• Any way to reduce BGP convergence time?– Avoid exploring paths with the same failure?– Hybrids of path vector and link state?

HLP: Hybrid Link-state/Path-vector

• Assume hierarchical AS structure– Provider-customer relationships dominate– And some peer-peer edges– (Are we willing to cook in these

assumptions?)

• Hybrid of link state and path vector– Link state within a sub-tree– Path vector across peer-peer links

• Route on AS numbers– Rather than prefixes

Add New Features in an Overlay: Resilient Overlay Networks

Overlay Networks

Overlay Networks

RON: Resilient Overlay Networks

Premise: by building application overlay network, can increase performance and reliability of routing

Two-hop (application-level)Berkeley-to-Princeton route

application-layer router

Princeton Yale

Berkeley

http://nms.csail.mit.edu/ron/

RON Circumvents Policy Restrictions

• IP routing depends on AS routing policies– But hosts may pick paths that circumvent

policies

USLEC

PU Patriot

ISP

meMy home computer

RON Adapts to Network Conditions

• Start experiencing bad performance– Then, start forwarding through intermediate host

A

C

B

RON Customizes to Applications

• VoIP traffic: low-latency path• Bulk transfer: high-bandwidth path

A

C

B

voice

bulk transfer

How Does RON Work?

• Keeping it small to avoid scaling problems– A few friends who want better service– Just for their communication with each other– E.g., VoIP, gaming, collaborative work, etc.

• Send probes between each pair of hosts

AC

B

How Does RON Work?

• Exchange the results of the probes– Each host shares results with every other

host– Essentially running a link-state protocol!– So, every host knows the performance

properties

• Forward through intermediate host when needed

AC

BB

RON Works in Practice

• Faster reaction to failure– RON reacts in a few seconds– BGP sometimes takes a few minutes

• Single-hop indirect routing– No need to go through many intermediate

hosts– One extra hop circumvents the problems

• Better end-to-end paths– Circumventing routing policy restrictions– Sometimes the RON paths are actually

shorter

RON Limited to Small Deployments

• Extra latency through intermediate hops– Software delays for packet forwarding– Propagation delay across the access link

• Overhead on the intermediate node– Imposing CPU and I/O load on the host– Consuming bandwidth on the access link

• Overhead for probing the virtual links– Bandwidth consumed by frequent probes– Trade-off probe overhead vs. detection speed

• Possibility of causing instability– Moving traffic in response to poor performance– May lead to congestion on the new paths

Future Routing Architecture

• Who is in charge?– Network administrators?– End hosts?– Third-party overlays?– Third-party routing providers?

• Build on top of today’s network?– New AS-level control plane?– Overlays on top of existing Internet?

• Assume (restricted) economic models?– To improve scalability and convergence?