Internet Anycast: Performance, Problems and Potential Zhihao Li , - - PowerPoint PPT Presentation

▶
internet anycast performance problems and potential
SMART_READER_LITE
LIVE PREVIEW

Internet Anycast: Performance, Problems and Potential Zhihao Li , - - PowerPoint PPT Presentation

Jun 25, 2023 326 likes •672 views

Internet Anycast: Performance, Problems and Potential Zhihao Li , Dave Levin, Neil Spring, Bobby Bhattacharjee University of Maryland 1 Anycast is increasingly used DNS root servers: All 13 DNS root servers Open DNS resolvers:

slide-1
SLIDE 1

Internet Anycast: Performance, Problems and Potential

Zhihao Li, Dave Levin, Neil Spring, Bobby Bhattacharjee University of Maryland

1

slide-2
SLIDE 2

Anycast is increasingly used

2

  • DNS root servers:
  • All 13 DNS root servers
  • Open DNS resolvers:
  • Google, Cloudflare, Quad9, OpenDNS, etc.
  • Content Delivery Networks:
  • Cloudflare CDNs (Stack Overflow,

Yelp, etc.)

slide-3
SLIDE 3

Mental model for anycast

3

  • Packets sent to an anycast address should travel to a

nearby site, subject to global/local constraints

  • More sites should mean lower latency, better distribution,

reliability against DoS attacks

slide-4
SLIDE 4

Anycast often chooses poorly

4

  • Actual distribution:
  • Many queries go to distant sites
  • Video below used queries from D-root on Dec. 1st 2016
slide-5
SLIDE 5

Outline

5

Do queries go to nearby anycast sites? Does adding anycast sites help? Performance Why not? Problems What can we do to fix it? Potential

slide-6
SLIDE 6

This is our data

6

  • Trace of DNS queries at D-root servers
  • ~20% of DNS queries to D-root at each site
  • Obtain approximate location of query source using MaxMind
  • Measurements from RIPE Altas probes
  • Over 9000 probes in ~180 countries and 3587 ASes
  • DNS CHAOS queries + traceroute
  • Measurements towards 9 out of 13 DNS root servers
slide-7
SLIDE 7

Do queries go to nearby sites?

7

  • We compute the extra distance each query travelled
  • ver their geographically closest site
  • Of course, we don’t expect clients to find the closest

site

  • But most queries should have short extra distance
slide-8
SLIDE 8

Over 1/3 queries traveled more than 1000km extra distance

8

The performance problem in D-root is representative of many current anycast deployments. Ideal

slide-9
SLIDE 9

Does adding anycast (global) sites help?

9

  • We analyze longitudinal RIPE Atlas measurements
  • Evaluate performance of 9 DNS roots in 2017
  • Compute average query distance in each week
slide-10
SLIDE 10

Site counts hardly matter

10

  • Each letter represents a performance value from one-week’s data
  • Vertical displacement of letters show performance variations

under the same number of site

slide-11
SLIDE 11

Site counts hardly matter

11

  • Each letter represents a performance value from one-week’s data
  • Vertical displacement of letters show performance variations

under the same number of site

slide-12
SLIDE 12

Outline

12

Do queries go to nearby anycast sites? Does adding anycast sites help? Performance Why not? Problems What can we do to fix it? Potential No! Over 1/3 queries traveled 1000+ km more than necessary; Adding sites hardly improves Performance Why not? Problems

slide-13
SLIDE 13

Why doesn’t anycast work as expected?

13

  • Routing and topology constraints introduce path

inflation

  • The actual path is longer than the shortest
  • However, path inflation in anycast is different…
  • Directs packets away from the closest anycast site
slide-14
SLIDE 14

14

Path inflation in unicast is like taking a detour

slide-15
SLIDE 15

Path inflation in unicast is like taking a detour

15

slide-16
SLIDE 16

Path inflation in anycast is different

Say we jump on a flight to “Cambridge”…

slide-17
SLIDE 17

17

Path inflation in anycast is like flying to a different Cambridge

slide-18
SLIDE 18

Anycast & unicast path inflation

18

Geographically closest Probe Lowest latency Selected
 anycast Unicast
 inflation Anycast
 inflation

Unicast inflation: Difference between the predicted latency[1] to geographically closest site, and latency to lowest latency site. Anycast inflation: Difference between latency to lowest latency site, and latency to selected site.

[1] Agarwal et al. Matchmaking for Online Games and Other Latency-Sensitive P2P Systems. SIGCOMM’09.

slide-19
SLIDE 19

Client A Client B Site A Site B

Anycast 199.7.91.13 Anycast 199.7.91.13 Unicast 10.1.1.1 Unicast 10.3.3.3

Use unicast representatives to estimate alternate site performance ≈

slide-20
SLIDE 20

Anycast path inflation is larger than unicast path inflation

20

D-root

Unicast Anycast

Cumulative Fraction

  • f probes

Number of probes

Ideal

slide-21
SLIDE 21

Anycast path inflation is larger than unicast path inflation

21

C-root D-root K-root

Unicast Anycast

All C-root sites share the same provider All other roots use multiple providers

slide-22
SLIDE 22

Would the best routes comply with typical routing policies?

  • Extract AS-level paths[2] from traceroutes
  • to selected site and to lowest-latency site
  • Find the ‘decision point’ where they diverge

[2] Mao et al. Towards an accurate AS-level traceroute tool. SIGCOMM’03

Probe Lowest latency Selected
 anycast Decision point

slide-23
SLIDE 23

Would the best routes comply with typical routing policies?

  • Extract AS-level paths[2] from traceroutes
  • to selected site and to lowest-latency site
  • Find the ‘decision point’ where they diverge
  • Examine if the better route is not selected due to:

[2] Mao et al. Towards an accurate AS-level traceroute tool. SIGCOMM’03

Prefer-Customer (PC) Shortest AS path (Short)

AS prefers routes through its customer ASes over the peer ASes, over its provider ASes AS prefers routes with shortest AS path length

slide-24
SLIDE 24

24

PC only PC+Short Anycast inflation

Can correct while following typical routing policies

D-root

Number of probes

slide-25
SLIDE 25

25

Can correct while following typical routing policies

D-root K-root

PC only PC+Short Anycast inflation

slide-26
SLIDE 26

Outline

26

What can we do to fix it? Potential No! Over 1/3 queries traveled 1000+ km more than necessary; Adding sites hardly improves Performance Why not? Problems What can we do to fix it? Potential Given routes with equal preference to different sites, BGP usually chooses poorly Problems

slide-27
SLIDE 27

Embedding geographic information in BGP announcements

27

  • Embed geographic location of the anycast sites

reachable through the announcement

  • Use BGP community attributes
  • Two 16-bit values X:Y
  • X represents the AS number that sets the

community

  • Y encodes the latitude and longitude
slide-28
SLIDE 28

Evaluate the geographic hint through simulation with real network traces

28

  • Using the traceroutes from each RIPE Atlas

probe

  • to the selected site and to lowest-latency site
  • Identify the ‘decision point’
  • Identify the geo-closest site to the ‘decision

point’, refer to as geo-hinted site

  • Benefits of geo-hint is the difference between

latency to geo-hinted site and to selected site

slide-29
SLIDE 29

A simple geographic hint provides large improvement on anycast performance

29

D-root

Fixed inflation Anycast inflation

Number of probes

slide-30
SLIDE 30

A simple geographic hint provides large improvement on anycast performance

30

D-root K-root

Fixed inflation Anycast inflation

slide-31
SLIDE 31

Outline

31

What can we do to fix it? Potential No! Over 1/3 queries traveled 1000+ km more than necessary; Adding sites hardly improves Performance Why not? Problems What can we do to fix it? Potential Given routes with equal preference to different sites, BGP usually chooses poorly Problems

A simple geo-hint in BGP communities can recover much

  • f the performance deficit

Potential

slide-32
SLIDE 32

Other results

  • Anycast does not distribute traffic in a balanced manner
  • Performance problems in anycast are common across

deployments

  • RIPE-Atlas measurements overrepresent Europe, but this

effect only over-estimates anycast performance

  • Unicast management addresses for anycast sites are good

representatives

  • Customized community attributes are more effective than

expected, given the default configuration in most routers

32

slide-33
SLIDE 33

Anycast doesn’t work as well as it should, but it can

  • Inefficiencies in anycast are excessive
  • Queries to most DNS roots travel to distant sites
  • Adding sites hardly improves anycast performance
  • Poor route selection in BGP causes larger path inflation in

anycast than in unicast

  • There exist equal-preference routes to closer sites
  • But no mechanism to choose the best among them
  • Incrementally deployable “geo-hints” in BGP can recover

most of performance deficit

cs.umd.edu/projects/droot/

Data available at:

33