Benchmarking BGP: Convergence in the Control Plane

1
Benchmarking BGPConvergence in the Control
Plane
2
Cooperative IETF Effort

• Marianne Lepp, Juniper Networks
• Howard Berkowitz, Gett Communications
• Elwyn Davies, Nortel Networks
• Susan Hares, NextHop
• Padma Krishnaswamy, Consultant
• Alvaro Retana, Cisco

3
IETF BMWG

• Juniper Networks, Inc.
• Expand discussion

• Benchmarking Methodologies Working Group
• Recommendations on measurements of performance of
  internetworking devices.
• Forum for discussion on the advancement of
  measurements.
• Benchmarking, not modeling or fishing trips.

4
Agenda

• Scope
• Define terminology
• Bound methodology
• Route mixture and Packet Trains
• Input desired

5
Terminology
Entire Internet
Entire AS
Router
6
Convergence Internet-Wide
Flapping
AS
AS
Hes down
AS
AS
AS
AS
7
Convergence Within an AS
Single AS
R
R
R
8
Convergence Single Box
Single Box
Tester
Tester
Routing/Control
Forwarding
9
Why Control Plane Convergence?

• Breaks complex problem into simpler problems
• Definable benchmark for comparisons
• Identifies bottlenecks within routing code
• Network problems often solved by looking into the
  box

10
Initial Draft

• Keep it Simple (Keep it Do-able)
• Single box
• Control Plane
• Black box testing
• No peeking under the covers (e.g. logs)
• eBGP only
• Works for IPv4/6

11
Topology Single Device Under Test
TR1
TRrx
DUT
Router
Traffic collector
TR2
Traffic Generators Peering neighbors

TRn
Synchronize clocks in measuring devices!!
12
Further Narrowing Scope

• iBGP specific Route Reflectors
• Confederations
• Communities
• Services
• RFC 2547 L3VPNs
• Multicasting
• Multi-protocol BGP
• Route flap damping
• Smoothing
• TCP

In V1?
13
Narrowing scope, cont.

• Inside the box
• Internal timers
• Policy, filtering, and authentication

In V1?
14
Test Requirements
15
Good testing

• Must be repeatable
• Must be consistent
• Must be specifiable
• Must take key factors into account
• BGP-specific functionality
• Peers types and connectivity
• Good data
• Route mixtures

16
Our Goals

• Baseline of expected performance
• Test different vendors fairly
• Design tests that can be replicated

17
Route MixturesWe need your help
18
Collected Data?

• Highly dependent on vantage point
• Tier 1 ISP, Enterprise, Access, etc.
• Looking Glass single view
• Scaling
• Test tables larger than existing tables
• Repeatable, consistent, and specifiable
• Could save packet traces

19
Modeled Data?

• Define parameters that characterize a route
  mixture
• Current work
• RTG, Saarland University in Germany
• Some work in Europe, not yet public

20
Prefix Distribution
Analyze current Internet tables to determine
prefix distribution characteristics
21
Node Distribution

• Example 2 sets of Class A /32 addresses

22
Node Distribution

• Different routers organize route database
  differently
• Need unbiased data, randomized route mixtures

23
BGP Attribute Distribution

• Prefixes that share an attribute are not
  necessarily grouped together
• Analysis shows an average of 2 consecutive NLRI
  share the same attributes
• 1.0.0.0/8 AS_PATH 100 200
• 2.0.0.0/8 AS_PATH 100 200
• 3.0.0.0/8 AS_PATH 200 300
• 4.0.0.0/8 AS_PATH 200 300
• 5.0.0.0/8 AS_PATH 200 300
• 6.0.0.0/8 AS_PATH 100 200

• Juniper Networks, Inc.
• (nexthop, communities, MED, localpref)

24
BGP Attribute Distribution

• Juniper Networks, Inc.Prossessing is the
  issueTry to combine slide make anal a table
  put to rightTable on next slide comes first

• Many attribute combinations possible
• Sample analysis

25
Packet Packing
26
Update Sequencing

• Parameters are
• Number of packets in each train
• Interval between packets
• Study/analyze packet traces
• Affected by TCP parameters, link traffic, BGP
  implementations

Packet 1
Packet 2
Packet 4
Packet 5
27
Key Parameters

• Prefix distribution
• Node distribution storage technique neutral
• AS mixtures and path lengths
• Attribute distribution
• Packet packing
• Packet arrival sequence

28
Conclusion on Route Mixtures

• Use Collected data, Use Modeled data
• But specify your data and how you got it.
• Work to be done on characterization
• Both composition and timing
• WE NEED Operator DATA
• WE NEED Operator INSIGHT

29
Summary

• BMWG cross-company effort
• Convergence inside the box answers interesting
  questions
• Narrowed our scope for version 1 to make it
  do-able
• We plan to build on this work

30
Summary cont.

• Route mixture must be specified clearly
• Identifying key parameters in work plan
• Modeled data changes with network and new
  understanding
• We want your help.

31
References
32
Much Work Outside the BoxA Sampling

• Ahuja and Labovitz, et al
• At Universities,
• Saarland in Germany, RTG
• UCLA, with Lixia Zhang
• Caida, www.caida.org
• Ripe, www.ripe.net
• Vendors, but theyre not telling

33
Outside the box, continued

• Stability and growth watchers IPMA, Tony Bate,
  Philip Smith, Geoff Huston
• IPMA http//www.merit.edu/ipma/
• Tony Bates site http//www.employees.org/tbates
  /cidr-report.html
• Phil Smiths stats www.apnic.net/stats/bgp/

34
References

• IETF51 BMWG talks
• www.ietf.org/proceedings/01aug/index
• .html
• under the ops area, bmwg
• BGP Convergence Terminology I-D
  http//www.ietf.org/internet-drafts/draft-ietf-bmw
  g-conterm-00.txt
• BGP Convergence Methodology I-D
  http//www.ietf.org/internet-drafts/draft-ietf-bmw
  g-bgpbas-00.txt

35
References Cont.

• Problematic Inter-domain Routing Issues, by Olaf
  Maennel, Anja Feldmann, Saarland University,
  http//www.nanog.org/mtg-0202/bgp.html
• See also http//www.net.unisb.de/olafm
• An Experimental Study of Delayed Internet Routing
  Convergence, Abha Ahuja, Farnam Jahanian, Abhijit
  Bose, Craig Labovits, RIPE 37 - Routing WG.
  http//www.ripe.net/ripe/meetings/index.html and
  a number of other papers.

36
Contacts

• mlepp_at_juniper.net
• hcb_at_gett.com
• elwynd_at_nortelnetworks.com
• skh_at_nexthop.com
• kri1_at_earthlink.com
• aretana_at_cisco.com

37
Thank You

• Questions?

38
Glossary

• AS Autonomous System
• BGP Border Gateway Protocol
• BMWG Benchmarking Methodology Working Group of
  the IETF
• DUT Device Under Test
• eBGP/iBGP Exterior/Interior BGP
• IETF Internet Engineering Task Force
• NLRI Network Layer Reachability Information
• RIB Routing Information Base
• TCP Transport Control Protocol
• TRi/TRrx Traffic Router, Receiving Traffic Router