Dynamics of HotPotato Routing In IP Networks

1
Dynamics of Hot-Potato Routing In IP Networks

• Renata Teixeira Aman shaikh Tim Griffin
  Jennifer Rexford
• Presented by Feng Wang
• Umass ECE

2
Introduction

• End-to-end Internet performance depends on the
  stability of the underlying routing protocols.
• Two-tiered routing architecture intradomain
  (OSPF or IS-IS) and interdomain (BGP) routing
  protocols.
• Many works focus on instability of BGP.
• Root cause and location of BGP dynamics.
• Impact of BGP dynamics on traffic.
• Little is known about how intradomain protocols
  influence BGP dynamics.

3
The interaction between IGP and BGP

• A router combines the BGP and IGP information to
  construct a forwarding table.
• BGP exchanges route advertisements with
  neighboring domains, and propagate reachability
  information within AS.
• IGP protocol, such as OSPF, computes shortest
  paths based on configurable link weights.
• The interaction between IGP and BGP
• Hot potato routing.

4
What is Hot potato routing?
destination
Neighbor AS
Egress points
B
C
9-gt11
10
A
AS1
Path cost change

• Hot potato routing direct traffic to the closest
  egress point the router with the smallest
  intradomain path cost, for example, router C.

5
Hot potato BGP routing changes
When the BGP decision process comes down to the
IGP Path cost, we refer to the BGP decision as
hot potato routing.
6
The impact of hot potato routing changes on
Internet performance

• Transient packet delay and loss while the routers
  recompute their forwarding tables.
• Shifts in traffic that may cause congestion on
  new paths.
• BGP routing changes visible to neighboring
  domain.

7
Is it easy to infer Hot-potato routing?

• Incomplete measurement data
• Latency
• BGP routing messages exchanged among routers
• OSPF flooded link-state advertisement.
• Only best path from BGP
• Complex routing protocol dynamics
• Due to receive several BGP or IGP updates, it is
  difficult to identify which BGP routing changes
  are caused by hot-potato routing.

8
Challenge of inferring Hot-potato routing

• Hierarchy of iBGP sessions inside the AS
• Vendor implementation details

9
Two approaches to infer hot-potato routing

• In this paper
• Controlled experiment black box
• Analysis of IGP and BGP measurements collected
  from a large ISP network.

10
Controlled Experiments

• Testbed

11
Results from controlled test
12
Measurement methodology

• Steps to correlate BGP updates with OSPF LSAs
• Compute the path cost vector from the OSPF LSAs
• Classification of BGP routing changes in terms of
  possible causes
• Associate BGP routing changes with related path
  cost change that occur close in time.

13
Compute cost vector changes

• Cost vector represents the cost of the shortest
  IGP path to every router in the AS
• Only consider those LSAs that reflect path cost
  changes
• Grouping path-cost changes that occur within a
  small time window into a single cost vector
  change. (the window size is 10 seconds from
  Figure 8 in the paper)

14
Classifying BGP routing changes

• Grouping BGP updates at the same router for the
  same prefix that occur close together in 70
  seconds.
• Narrow down possible causes

15
Matching Cost Changes with BGP

• Matching the cost changes with BGP changes within
  a time window (-2, 180).

16
Extent of hot potato BGP changes
17
Recommended operational practices

• Selection of IGP path costs to egress points.
• Traffic engineering and planned maintenance.

18
conclusions

• Hot-potato routing can be a significant source of
  BGP updates.
• BGP updates can lag 60 seconds or more behind the
  intradomain event.
• The fraction of BGP messages triggered by
  intradomain changes varies significantly across
  time and router locations.