Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha,...
Transcript of Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha,...
![Page 1: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/1.jpg)
11
Measuring Reverse Paths
Ethan Katz-Bassett, Harsha V. Madhyastha,Arvind Krishnamurthy, Thomas Anderson
University of WashingtonAugust 2008
This work partially supported by Cisco, Google, NSF
![Page 2: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/2.jpg)
2
Reverse Paths Would Be Useful
Many distributed systems would benefit fromreverse path information
Hubble to isolate failures and group problems iPlane, Path-Stitching to provide more
accurate path and property predictions Ark, etc., for more complete topologies Google to find inflated paths back from clients ISPs to find inflated paths back to customers
![Page 3: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/3.jpg)
3
Current Tools Don’t Provide That Info ping, traceroute
Simple tools proven useful for many systems Only provide forward path or round-trip info
Existing one-way tools require control of bothends RIPE’s TTM infrastructure owping
Vantage points could solve problems Prober in every home?
![Page 4: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/4.jpg)
4
Goals
Techniques for reverse traceroute andone-way ping when we do not control destination
Evaluate how often they work Demonstrate how they help us understand Internet
Systems from earlier slide: iPlane, topology, Google Asymmetry Daily reverse map from world back to PlanetLab
Preliminary/ongoing for nowTalk will focus on reverse traceroute
![Page 5: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/5.jpg)
5
Reverse Traceroute Approach
Exploit destination-based routing IP options carried over to response packets
Timestamp option (TS): time-query 4 ordered IPs Record route option (RR): first 9 routers recorded
Spoofing to overcome: Lack of vantage points in most prefixes Max 9 hops recorded with RR Limited support/ filtering of options
![Page 6: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/6.jpg)
6
Spoofing?? Isn’t that bad? We use only a restricted version
Only spoofing as nodes we control Rate limit, restrict destinations (no broadcast IPs)
Millions of spoofed probes sent to 10s ofthousands of IPs, no complaints
Hubble and this work show utility Lets us approximate:
Having control of destinations One-hop detouring/ loose source routing One VP sending to another, bouncing through dst
![Page 7: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/7.jpg)
7
Want reverse path from D back to S, but don’t control D Set of vantage points, some of which can spoof
![Page 8: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/8.jpg)
8
Traceroute from all vantage points to S Gives atlas of paths to S; if we hit one, we know rest of path
![Page 9: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/9.jpg)
9
To: DFr: SPing?TTL=8
To: DFr: SPing?TTL=0
To: DFr: V3Ping?TTL=8
To: DFr: V3Ping?
To: V3Fr: DPing!
To: V3Fr: DPing!
From all vantage points, ping D with TTL=8 to find thosewithin 8 hops
Record route does 9 hops, so these will give us return hop(s)
![Page 10: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/10.jpg)
10
To: DFr: SPing?RR:__
To: DFr: SPing?RR: h1,…,h7
To: SFr: DPing!RR: h1,…,h7,D
To: SFr: DPing!RR: h1,…,h7,D,R1
From vantage point within 8 hops of D, ping D spoofing as Swith record route option
D’s response will contain recorded hop(s) on return path
![Page 11: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/11.jpg)
11
To: R1Fr: SPing?RR:__
To: SFr: R1Ping!RR: h1,…,h6,R1,R2,R3
Iterate, performing TTL=8 pings and spoofed RR pings foreach router we discover on return path
![Page 12: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/12.jpg)
12
![Page 13: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/13.jpg)
13
To: R3Fr: S Ping?TS: R3? R4?
To: SFr: R3Ping!TS: R3! R4?
To: SFr: R3Ping!TS: R3! R4!
If no spoofing vantage points within 8 hops, consider set ofrouters directly connected to R3 (in pre-measured topology)
Use timestamp option to try to verify which is on return path
![Page 14: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/14.jpg)
14
![Page 15: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/15.jpg)
15
Once we see a router on a known path, we know remainder
![Page 16: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/16.jpg)
16
Techniques combine to give us complete path We have additional techniques for inferring reverse hops
![Page 17: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/17.jpg)
17
Preliminary results
Spoofing gives a few extra hops to connect to measuredpaths
End hosts like PL are a few extra hops from routers PL-PL measurements more likely to share paths (GREN)
Reverse paths from PLsites back to UW
Measurements: TR PL to UW RR PL to UW Spoofed RR as UW Pick dst, exclude site
How many hops backfrom dst need to be givenbefore we can constructa complete path for restof reverse TR?
Median: 3 vs 5
![Page 18: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/18.jpg)
18
Reverse Path Summary Reverse path info can be very useful to systems Ongoing work on reverse traceroute and
one-way ping for when we don’t control destination Preliminary results here and in Hubble show
techniques can work
Limiting factors: Restricted support for options Current prober deployment
Need diverse paths back to our test sources Need spoofing vantage points in diverse network locations Any we can use?
![Page 19: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/19.jpg)
19
Measurement Work at UW
Real Internet-scale measurement-basedsystems Hubble - Monitoring black holes on the Internet iPlane - Providing Internet path and path property
predictions Ongoing work
Reverse path techniques Massive software prober deployment Evaluating prober deployments
![Page 20: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/20.jpg)
20
Massive Prober Deployment
Goal: on-demand probes from any prefix Talking with RIPE Science Group about 3 tier
brain/ controller/ prober architecture Different classes of probers operating under
standard controllers Super probers - TTM, PlanetLab Hardware probers - simple USB dongles Software probers - next slide
![Page 21: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/21.jpg)
21
Software Probers: Incenting End-users
Plan to develop software prober plugin Deploy in different vehicles that incent users
to contribute measurements by providingbenefit of measurements BitTorrent client Reliability-focused detouring - Firefox plugin Apps built on iPlane predictions
![Page 22: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/22.jpg)
22
Measurement Work at UW
Real Internet-scale measurement-basedsystems Hubble - Monitoring black holes on the Internet iPlane - Providing Internet path and path property
predictions Ongoing work
Reverse path techniques Massive software prober deployment Evaluating prober deployments
![Page 23: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/23.jpg)
23
Reviewers (properly) suspicious of PLActual (paraphrased) comments from reviews: “Needs evaluation of likely coverage of all paths in
Internet given small size of PlanetLab” “Let me know how much of Internet is observable
and suggest vantage points to improve coverage” “Oddities of Abilene are hard to reason about” “Including more text on limitations of PlanetLab” “Include discussion on how well you see this
technique working in the global Internet.”
![Page 24: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/24.jpg)
24
Assessing prober deployment
Previous work either focuses on: Measurements between vantage points Cumulative topology
Our focus: Paths to prefixes
![Page 25: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/25.jpg)
25
Goals
Techniques to help with: node selection for a system: # and which node deployment: where to place new nodes assessing how set of vantage points
represents overall diversity of paths and howresults of a study would vary with a differentdeployment
![Page 26: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/26.jpg)
26
Questions to answer
Ideal is every end host. How close is ourdata to that? How much does spoofing help?
Is PlanetLab limited primarily by # of sites oralso by network locations of sites?
How many vantage points do we "need?" How much does it help to select vantage
points per target vs one set for all targets? How can we characterize which nodes are
most useful to add?
![Page 27: Measuring Reverse Paths - CAIDA1 Measuring Reverse Paths Ethan Katz-Bassett, Harsha V. Madhyastha, Arvind Krishnamurthy, Thomas Anderson University of Washington August 2008 This work](https://reader030.fdocuments.in/reader030/viewer/2022040812/5e568a53c0e95c76653ae1fd/html5/thumbnails/27.jpg)
27
Measurement Work at UW Real Internet-scale measurement-based
systems Hubble - Monitoring black holes on the Internet iPlane - Providing Internet path and path property
predictions Ongoing work
Reverse path techniques Massive software prober deployment Evaluating prober deployments
Would love to talk about or collaborate on anyof this.