ASAP: an AS-Aware Peer-Relay Protocol for High Quality VoIP

1
ASAP an AS-Aware Peer-Relay Protocol for High
Quality VoIP

• Shansi Ren, Lei Guo, and Xiaodong Zhang
• Ohio State University

2
VoIP Introduction

• VoIP vs. PSTN telephones
• VoIP includes signaling and voice packet
  transmission process
• Skype is a popular commercial P2P VoIP application

3
VoIP Packets Traveling in Internet
Long router queuing delay!
Alice, I am Bob
Bob, What did you say?!!
Internet
Alice
Bob
Bob -gt Alice voice pkts.
Internet routing is Critical for VoIP Quality!
Alice -gt Bob voice pkts.
4
VoIP Quality Requirements

• Mean Opinion Score (MOS) metric
• MOS gt 3.5 is acceptable.
• Network factors
• E2E one-way latency lt 150 ms
• E2E loss rate lt 0.5

5
Some Facts of VoIP in Internet

• Two end hosts communicate through their direct IP
  routing path by default.
• Direct routing path may not always meet the VoIP
  quality requirements.
• Overlay routing sometimes may have better routing
  performance.

6
direct, 1-hop, and 2-hop overlay routing
host B
host C
host A
direct IP routing path
1-hop overlay routing path
Internet
host D
2-hop overlay routing path
host E
host A communicates with host C
7
Internet Routing

• Internet consists of Autonomous Systems (ASes),
  and hosts are administered in the unit of AS.
• ASes are connected by core routers, and routing
  between ASes relies on the Border Gateway
  Protocol (BGP).
• Connected ASes has customer-provider and
  peer-peer relationship.
• An customer AS connecting to multiple upstream
  ASes is called a multihoming AS. (or multiple
  providers)
• Valley-free a direct Internet routing path has
  the form (customer-provider)(peer-peer)?(provider
  -customer).

8
AS Connection and Multihoming
peer-peer connection
provider AS 1
provider AS 2
customer AS is multihomed
customer-provider connection
customer AS
9
An Internet e2e Path is Valley-free
AS D
AS B
AS C
valley-free path
AS B
AS C
AS A
AS D
path is NOT valley-free This is an impossible
direct path!
AS B
AS A
AS A
AS C
provider-to-customer edge
peer-to-peer edge
valley-free path
10
Targeted Research Questions

• How insufficient is Internet direct routing for
  VoIP?
• Under what condition, can overlay routing (relay)
  improve VoIP quality?
• What kind of quality does Skype provide, and what
  are the limits in its routing?
• How to design efficient routing methods for high
  quality VoIP with low overhead?

11
Outline of Talk

• VoIP application introduction
• Internet e2e latency measurements
• Skype measurement and observations
• ASAP protocol design and evaluation
• Conclusion

12
E2E Latency Measurement Procedures
Gnutella IP probing
online Gnutella IP addresses
Limewire software modification
Gnutella IP crawler
BGP tables and updates
IP prefix and origin AS extraction
IP prefix table
AS-level cluster Identification and delegate
IP selection
cluster delegate IP addresses
pairwise IP DNS server latency measurement
pairwise delegate IP latency
King tool based prober development
King prober
13
Sessions and their RTTs

• A session consists a pair of end host.
• We randomly generate 105 sessions among cluster
  delegates.
• We measure session direct RTTs using king
  facility.
• For delegates a, b, c, relay path a-b-c
• RTTa-b-c RTTa-b RTTb-c relay delay.

14
Internet e2e RTT Measurement
DNS server of host c
DNS server of host b
IP of host c?
host b
IP of host c
Internet
IP of host c
host c
IP of host c?
host a
host a measures RTTb-c via recursive DNS queries
15
Direct vs 1-Hop RTT
50 sessions have optimal 1-hop RTT lt direct IP
RTT
25 sessions whose opt. 1-hop relay can reduce
direct IP RTT by more than 50
16
Overlay Routing Reduces RTT
Sessions whose direct IP RTTs gt 300 ms
Sessions opt. 1-hop RTTs are always lt 300 ms
17
Relay Improves VoIP Quality

• There are 2 and 10 of sessions with direct RTTs
  above 300 ms and 250 ms, respectively.
• We can always find one-hop relay paths whose RTTs
  are below the threshold for these sessions.
• Peer relay plays an important and critical role
  in improving the quality for VoIP applications.

18
Direct Path Is Congested
direct path between AS A and AS C
AS H is congested
AS G
AS H
AS D
AS E
AS F
1-hop relay path between AS A and AS C via AS B
AS A
AS B
AS C
direct path between AS B and AS C
direct path between AS A and AS B
provider-to-customer edge
peer-to-peer edge
19
Multi-homed AS B As 1-hop Relay
direct path between AS A and AS C
AS H
AS I
AS F
AS G
AS D
AS E
AS B
AS C
AS A
direct path between AS A and AS B
direct path between AS B and AS C
1-hop relay path between AS A and AS C via AS B
AS B is multi-homed, connects to AS A and AS C
provider-to-customer edge
peer-to-peer edge
20
Outline of Talk

• VoIP application introduction
• Internet e2e latency measurements
• Skype measurement and observations
• ASAP protocol design and evaluation

21
Skype Experimental Sites and Sessions
Vancouver, Canada
Dalian, China
Jersey City, NJ
Bozeman, MT
Beijing, China
Reston, VA
Baltimore, MD
Shanghai, China
Williamsburg, VA
Jingzhou, China
Austin, TX
We have chosen 14 representative Skype sessions
22
Skype Relay Selection Limits

• Limit 1 Long latency due to improper relay node
  selections.

Session 4
300 ms
Session 10
300 ms
23
Limit 2 Probing multiple latent nodes in the
same AS.
two probed relay nodes in session 8
relay node DNS zone name relay
path RTT 85.64.x.x
barak-online.net 360 ms 85.65.x.x
barak-online.net 359 ms
Limit 3 Taking a long time to find major relays.
24
Limit 4 Generating non-negligible overhead.
before stabilization
10
10
after stabilization
25
Outline of Talk

• VoIP application introduction
• Internet e2e latency measurements
• Skype measurement and observations
• ASAP protocol design and evaluation

26
Skype Measurement Summary

• Although we do not know the routing algorithm of
  Skype
• Non-optimal replay nodes are used often.
• Seems to only reply on probes to find a relay
  node in a ad-hoc way many probes.
• The relay nodes are frequently changed even after
  the sessions are established.
• It is an AS-unaware routing.

27
ASAP AS-Aware Peer-Relay Selection Method
28
ASAP Design Rationale

• In general, peer nodes with the same IP prefix
  are relatively close to each other.
• With publicly available BGP tables and updates,
  an up-to-date annotated AS graph can be built.
• Paths with longer AS hops are likely to have
  longer latencies.
• An Internet AS-level direct IP routing path
  usually has the valley-free property.

29
Three Types of ASAP Nodes
bootstraps data structure
bootstrap1
bootstrap2
Internet AS graph
IP prefix to cluster Surrogate IP table
cluster surrogates data structure
IP prefix to ASN table
clusters close cluster set
Type 1 bootstraps
Type 2 surrogates
Internet AS graph
clusters top node table
surrogate SA
cluster C
surrogate SB
end host h3
cluster A
cluster B
end host h1
Type 3 end hosts
end host h2
30
Close Clusters Construction Process
AS 5
h6
h4
AS 4
AS 6
s5
s4
s6
good, 220 ms
good, 180 ms
s2
AS 2
h3
AS 3
bad, 350 ms
good, 75 ms
s3
s1 close cluster
s1
s2 75 ms
good, 52 ms
AS 1
s5 180 ms
ping
s6, h6 220 ms
h1
pong
s3, h3 52 ms
good RTT lt 300 ms loss rate lt 5
provider-to-customer edge
peer-to-peer edge
bad RTT gt 300 ms loss rate gt 5
31
h1-h4 Close Relays Selection Process
AS 5
h6
h4
AS 4
AS 6
s5
s4
s6
s4 close cluster
s2 50 ms
s2
AS 2
h3
s5 170 ms
AS 3
s3
s1
RTTh1-s2 RTTh4-s2 125 ms lt 300 ms
s1 close cluster
s2 is good relay for h1-h4 VoIP session
AS 1
s2 75 ms
h1
s5 180 ms
s6, h6 220 ms
s3, h3 52 ms
provider-to-customer edge
peer-to-peer edge
32
ASAP Call Session Process
bootstrap1
bootstraps data structure
bootstrap2
cluster surrogates data structure
Internet AS graph
IP prefix to cluster Surrogate IP table
clusters close cluster set
surrogate?
IP prefix to ASN table
Internet AS graph
surrogate IP
clusters top node table
surrogate?
surrogate SA
cluster C
surrogate IP
surrogate SB
voice pkts
voice pkts
close set?
voice pkts
close set
end host h3
voice pkts
close set?
close set
cluster A
cluster B
end host h1
h2s close set
end host h2
control pkt.
h2s close set?
voice pkt.
33
Evaluation Metrics

• Number of quality paths number of relay paths
  satisfying the RTT and loss rate requirements
• Shortest RTT and highest MOS of these quality
  paths
• Overhead measured by the number of generated
  messages to find quality path relay nodes

34
Different Routing Methods

• DEDI uses dedicated relay nodes. (SOSP01)
• RAND randomly selects relay nodes. (OSDI04)
• MIX is a combination of RAND and DEDI.
• ASAP selects relay nodes using our AS-aware
  method.
• OPT always chooses relay nodes that give the
  shortest overlay routing latency. (Offline method)

35
Number of Quality Paths
For 90 sessions, ASAP can find more than 5,000
quality paths
DEDI, RAND, and MIX can find no more than 500
quality paths for all sessions
36
Shortest Path RTT
115 ms
1 s
In DEDI, RAND, and MIX, more than 5 sessions
have shortest RTT gt 1s
In ASAP and OPT, all sessions have shorest RTT lt
115 ms
37
Highest Path MOS
2.9
3.85
In DEDI, RAND, and MIX, about 3 sessions have
highest MOSs lt 2.9
In ASAP and OPT, all sessions have highest MOSs gt
3.85
38
ASAP Is Highly Scalable
23,366 end hosts
103,625 end hosts
The number of quality paths found by ASAP remains
stable under different end host population.
39
ASAP Has Moderate Overhead
In ASAP, 85 sessions generate less than 300
messages
DEDI, RAND, and MIX all probe fixed number of
nodes, i.e., 160, 160, and 200 nodes
40
Conclusion

• In a global overlay systems, 10 sessions of
  direct path cannot meet VoIP quality
  requirements.
• For these sessions, there always exist multiple
  relay paths that can meet the requirements.
• Existing relay selection methods, including
  Skype, do not always select proper relay nodes.
• Optimal replay nodes can be found by AS-aware
  routing.
• We show ASAP is scalable, light-weight, and
  outperforms all existing solutions.

41

• ASAP source code and results can be found at
• http//www.cse.ohio-state.edu/sren/VoIP-Peer-Rela
  y/