Transcript Drafting Behind Akamai (Travelocity

Drafting Behind Akamai
(Travelocity-Based Detouring)
Dr. Yingwu Zhu
Motivation
• Overlay networks
– Build the “view” of the underlying network
– Passive and active measurements
• To determine overlay paths according to some metric,
e.g., latency?
• Redundant, non-scalable, & overhead (expensive!)
• What common service we may need in place?
– An Internet weather-report service
Proposed Approach
• Fact
– CDNs (e.g., Akamai) perform extensive network and
server measurements
• Publish the results through DNS over short time scales
• Can overlay networks recycle measurements
collected by production CDNs?
– Significantly reduce the amount of measurements (a
complementary service)
– No new infrastructure need to be deployed
– Inherit the robustness of DNS
– Easy integration with existing systems
CDN-Driven One-Hop Source Routing
D
A1
E1
A2
E2
An
En
S
DNS Server
Key Questions
• How does Akamai work?
–
–
–
–
–
DNS translation
How many web replicas does a client see?
Impact of different sites (e.g., Yahoo vs. NY Times)?
DNS redirection dynamics?
Network or server latency?
• An example application: one-hop source routing
– Mapping CDN servers to overlay nodes
– Low-overhead protocols for exploiting CDN
redirections
DNS “Black Magic”
Root DNS Server
Akamai High-Level
DNS Server
Top-Level Domain
DNS Server
www.pcworld.com
www.pcworld.com
a1694.g.akmai.net
CNAME:
images.pcworld.com
www.pcworld.com
a1694.g.akamai.net
Akamai Low-Level
PCWorld Authoritative
DNS Server
a1694.g.akmai.netDNS Server
2 ip addresses of
Local DNS Server Akamai Edge Servers
images.pcworld.com
PCWorld
Web Server
6
2 ip addresses of
www.pcworld.com
Akamai Edge Servers
Akamai
Edge Server
Web
Client
Measuring Akamai
• 2-months long measurement
• 140 PlanetLab (PL) nodes
– 50 US and Canada, 35 Europe, 18 Asia, 8 South America, the rest
randomly scattered
• Every 20 sec, each PL node queries an appropriate CNAME for
– Yahoo, CNN, Fox News, NY Times, etc.
Akamai
Edge Server 1
Akamai Low-Level
DNS Server
7
PL Node
Akamai
Edge Server 2
.……
Akamai
Edge Server 3
Initial Results
Berkeley
Purdue
day
night
8
Server Diversity for Yahoo
Majority of PL nodes
see between 10 and 50
Akamai edge-servers
Nodes far away
from Akamai
hot-spots
9
Good overlay-to-CDN
mapping candidates
Multiple Akamai Customers
Yahoo and NYTimes are
not hosted in U. Oregon
Amazon is not hosted
on UMass and LBNL
FEMA is poor!
10
Redirection Dynamics
Berkeley
Brazil
Korea
11
Key Questions
• How does Akamai work?
–
–
–
–
–
DNS translation
How many web replicas does a client see?
Impact of different sites (e.g., Yahoo vs. NY Times)?
DNS redirection dynamics?
Network or server latency?
• Potentials for one-hop source routing?
– Mapping CDN servers to overlay nodes
– Low-overhead protocols for exploiting CDN
redirections
12
Methodology
10 Best
Akamai
Edge Servers
………
Akamai Low-Level
DNS Server
13
Planet Lab Node
Do CDN redirections correlate to network
latencies? (1)
• Rank = r1+r2-1
– 16 means perfect correlation
– 0 means poor correlation
Brazil is poor
14
MIT and Amsterdam
are excellent
Do CDN redirections correlate to network
latencies? (2)
70% paths are
among best 10%
97% better than
average
CNN is poor!
15
Akamai-Driven One-Hop Source Routing
Redirections driven by
network conditions
D
Potential for
CDN-to-overlay mapping
A1
E1
A2
E2
An
S
DNS Server
16
Redirection dynamics
sufficiently small for
network control
En
Methodology
Akamai path 1
Akamai path 2
Destination
Akamai path 3
Akamai path 1
Akamai path 10
Akamai path 2
Akamai path 3
Akamai Low-Level
DNS Server
Akamai path 10
Source
17
Direct Path
Akamai-driven source routing (1)
Taiwan-UK
80% Taiwan
15% Japan
5 % U.S.
18
UK-Taiwan
75% U.K.
25% U.S.
Akamai-driven source routing (2)
Experiment: US (6), Europe (3), S. America (2), Asia (3)
Direct paths better
than Akamai paths
19
Akamai and direct
paths have similar
performance
25% of Akamai
paths better than
direct paths
Path pruning
• Fact
– Not always is Akamai-driven path better than the
direct one
• Practical issues
– How frequently to make a decision whether to use
the direct or the Akamai path?
– Should one use
• the first (of the 2 paths) returned by Akamai (FAS)
• the better (of the 2 paths) returned by Akamai (BTAS)
20
Path Pruning Result
BTAS better than FAS
Direct path accounts for
78% of the gain
21
2 hours update frequency
before the performance declines
Conclusions
• Reuse measurements performed by CDNs
• Reverse-engineering Akamai
– DNS redirections sufficiently small
– Strong correlation to network conditions
– All clients see a large number of paths
• CDN-driven one-hop source routing
– 25% of Akamai paths outperform direct paths
– 50% of nodes “discovered” by Akamai outperform
direct paths
– Low-overhead pruning algorithms
• Global Internet “weather-report” service for little
to no cost
22