Paxos made simple

1
CS234/NetSys210 Advanced Topics in
Networking Spring 2012 SIP and VoIP Kundan
Singh, and Henning Schulzrinne "Peer-to-peer
internet telephony using SIP" in NOSSDAV
'05 Presentation by Swaroop Kashyap Tiptur
Srinivasa Anirudh Ramesh Iyer Tameem Anwar
2
Introduction to Session Initiation Protocol
(SIP)
3
Introduction to SIP

• What is SIP ?
• Text-based protocol (Defined in RFC 3261)
• SIP Applications
• Network Elements
• User Agent (UAC and UAS)
• Proxy Server (UAC and UAS)
• Registrar
• Redirect Server
• SBC

4
SIP Messages

• Request messages
• REGISTER
• INVITE
• ACK
• CANCEL
• BYE
• OPTIONS

5
SIP Messages

• Response messages
• PROVISIONAL (1XX)
• SUCCESS (2XX)
• REDIRECTION (3XX)
• CLIENT ERROR (4XX)
• SERVER ERROR (5XX)
• GLOBAL FAILURES (6XX)

6
Assessment of VoIP quality over Internet
backbones Athina P. Markopoulou, Fouad A. Tobagi,
Mansour J. Karam

• Quality of service for VoIP traffic in Internet
  backbone was studied based on delay as the
  metric.
• 7 ISPs, 43 paths formed the test setup.
• A model is proposed to study and analyze VoIP
  traffic in Internet backbones based on
  characteristics of VoIP such as talkspurts and
  silence periods.
• This model is applied to the data trace obtained
  from the test setup (2.5 days worth of data).
• The study shows that certain backbone paths are
  not equipped to handle VoIP traffic. Examples of
  such paths are coast-to-coast paths which have
  high delay.
• The authors suggest that the primary reason for
  high delay in paths is due to the fact that there
  is no distinction between data and voice traffic.
  This prompts the authors to suggest IP QoS
  mechanism as a solution.

7
Improving VoIP Quality thorugh Path Switching Shu
Tao , Kuai Xu, Antonio Estepa, Teng Fei ,Lixin
Gao ,Roch Guerin, Jim Kurose, Don Towsley,
Zhi-Li Zhang

• The effectiveness and benefits of path switching
  was examined, and its feasibility was
  demonstrated with the help of a of a prototype
  application-driven path switching gateway
• With sufficient path diversity, path switching is
  indeed capable of yielding meaningful
  improvements in voice quality
• The experiments also highlighted the benefit of
  adaptive decisions, especially in light of the
  often changing nature of the time scale at which
  network congestion takes place.
• The study suggests that by exploiting the
  inherent path diversity of the Internet,
  application-driven path switching is a viable
  option in providing quality-of-service to
  applications
• There is ongoing research being done to pursue
  these issues further in the context of hybrid
  wired/wireless networks and other applications
  such as video.

8
QoS-Enabled Voice support in the Next-Generation
Internet Issues, Existing Approaches and
Challenges Bo Li, Mounir Hamdi, Dongyi Jiang,
and Xi-Ren Cao, Hong Kong University of Science,
Technology ,Y. Thomas Hou, Fujitsu Laboratories
of America

• There has been significant work done to establish
  the foundation to support VoIP. However, much
  remains to be done in order to ensure the QoS for
  VoIP and for multimedia traffic in general.
• This article surveys the existing technologies
  to support VoIP, in particular the basic
  mechanisms in the IETF Internet telephony
  architecture and ITU-T H.323-related
  recommendations.
• It then reviews the IETF QoS framework and major
  components in providing such QoS guarantees,
  including the Intserv and Diffserv models.
• In addition, this article also presents two
  leading companies (Cisco and Lucent) solutions to
  offering IP telephony services
• One another major issue currently under active
  development is internetworking with legacy net-
  works (i.e., PSTN). There are a number of
  proposals within the IEFT, in particular
  Media Gateway Control Protocol (MGCP).

9
Peer-to-peer internet telephony using
SIP Kundan Singh, and Henning Schulzrinne,
NOSSDAV '05
10
Peer-to-Peer Internet Telephony using SIP

• SIP using Client-Server model
• Less Robustness and Scalability
• Increased costs due to Maintenance and
  Configuration
• SIP using Peer-to-Peer model
• Increased Robustness and Scalability
• No maintenance and Configuration
• Interoperability
• Tradeoff
• Resource look-up
• Security

11
Distributed Hash Table ( DHT)

• Types of Search
• Central Index (Napster)
• Distributed Index with flooding (Gnutella)
• Distributed Index with hashing (Chord)
• Basic Operations
• find(key),insert(key , value),delete (key),
• But no search

12
Background and Related work

• Chord
• Ring based Distributed Hash Table for structured
  P2P systems
• Identifier Circle
• Keys assigned to successor
• Evenly distributed keys and nodes
• Finger table- O(log N) entries
• ith finger points to first node that
  succeeds n by at least 2i-1
• Stabilization for join/leave
• Iterative and Recursive Routing

13
Background and Related work

• Skype
• Based on Kazaa architecture
• Open source P2P application for Internet
  telephony and instant messaging
• Uses the concept of Super nodes.
• Proprietary with Global Index server assigning
  Super nodes
• Lookup similar to Kazaa using flooding unlike
  DHT-based lookup
• Explicit server configurations not required

14
Background and Related work

• P2P with SIP
• Recent work have concentrated on combining SIP
  and P2P
• SIP combination with P2P done in two ways
• Replace SIP user registration and lookup by an
  existing P2P protocol
• Implementation a P2P algorithm using SIP
  messaging
• The paper takes the second approach
• No modification done to SIP messages
• Advantages- Use of existing SIP components
• Disadvantages- Transport message size overhead

15
Difference with File Sharing

• A single P2P-SIP node can handle many more
  requests than a file sharing node due to low data
  volume
• Caching of location information is not useful.
• The file sharing and directory lookup-based
  systems
• can tolerate high lookup latency.
• For file sharing applications, multiple almost
  exact copies of a popular file may be available.
  So node reliability does not matter.

16
Architecture

• Server Farm architecture
• P2P Overlay architecture
• Hybrid Super-Node architecture

17
Server Farm architecture

• Preserves Client-Server model
• DHT can used in a Server farm
• User registration done on only O(logN) servers
• Redundancy in servers can prove expensive

18
Client Overlay architecture

• Pure P2P overlay with all clients acting as a
  server
• No server maintenance and configuration needed
• Problem- Equal capacity and availability
• Example- Client behind firewall or NAT

19
Super Node architecture
/watch?v1pKztSTmeIc
/watch?v1pKztSTmeIc

• Hybrid design (Similar to Kazaa)
• Selection of Super nodes
• High capacity (Bandwidth, CPU, memory)
• Availability (up time, public IP address)
• Transition from a regular node to Super node
  (Local decision)

20
P2P-SIP node block diagram

• Discover-gtUser Interface-gtDHT(Chord)

21
Design and Implementation

• Naming
• Authentication
• SIP messages
• DHT discovery and join
• SIP message routing
• Reliability
• Adaptor for existing SIP phones

22
Naming

• Representation of end points using SIP URI's
• Eg- sip17_at_192.1.2.38054
• 17 is the key returned by Chord's hash function
• Domain representation encouraged
• sip10_at_example.com
• SIP user identifiers (UI) randomly assigned
• Authentication ?

23
Authentication

• Validity of UI done when an user signs up to the
  P2P-SIP network
• Absence of Public Key Infrastructure ?
• Password obtained used in REGISTER authentication
• Time-to-Live value used to determine user
  existence

24
SIP messages

• SIP REGISTER- Used for registration and DHT
  maintenance
• SIP REGISTER used in query and update mode
• Query mode- User is asking for Contact
  information of the node identifier given in TO
  header
• Update mode- Contact information present in
  message
• Mostly done to update bindings given in TO header

25
DHT Discovery and Join

• SIP REGISTER message used, Request-URI could be
  sip224.0.1.75
• TO header identifies and discovers other P2P-SIP
  peers in the network
• Once the node is discovered, SIP REGISTER is used
  to join the DHT
• Chord Stabilization ? (Hint Registration of end
  points)

26
SIP message routing

• Every node owns the responsibility for its subset
  of the key space
• Destination key is determined from TO header
• If Destination key belongs to same key space of
  the received node, then node is the registrar
• Redirect or the proxy server is used to determine
  the locations available for the destination user.
• Otherwise, request is proxied to the next hop
  node based on Chord algorithm and its inherent
  data structures

27
Reliability

• Each Chord node stores log(N) successor addresses
• User registrations replicated at K successive
  nodes
• Users unregistered when exit is graceful
• Registrar Node malicious ?

28
Advanced Services

• Apart from user registration and call routing,
  which are the core services of SIP, P2P-SIP also
  offers some advanced services.
• These services are based on SIP URIs. Example-
  sipstaff-meet_at_office.com or dialog.voicexml_at_ivr.
  net
• The services are
• Offline messages
• Multi-party conferencing
• NAT and firewall traversal
• Directory Service

29
Advanced Services Offline Messaging

• Offline messages Caller calls callee and callee
  is unavailable then a message is left by caller
  to callee. Needs storage and signaling.
• Current P2P file storage IP telephony cannot
  handle this as we also need message waiting
  indication.
• The P2P-SIP client running on the destination
  users machine can store the message if the
  destination user did not pick up the phone. The
  problem comes when the destination phone itself
  is not active or the user has not started her
  client.
• The message waiting signal is implemented using
  POST (POST is a cooperative, decentralized
  messaging system that supports traditional
  services like email, news, IM) which is built on
  a P2P overlay.
• To receive the offline message, the destination
  node subscribes to the message waiting indication
  (MWI) event with the P2P network and gets
  notified on startup when a new offline message is
  available .

30
Advanced Services Offline Messaging Contd

• There are three places where we can store the
  of?ine messages the source, the destination or
  some intermediate node in the P2P overlay.
• When Alice calls Bob and Bob is not online, the
  message should be stored reliably by the system
  and delivered to Bob when he comes online. Then a
  possible solution is to use the DHT peer
  responsible for storing Bob' location also store
  his offline messages.
• In case of storage node failure message is
  replicated and kept consistent using Oceanstore
  architecture (a well-known architecture for
  global-scale persistent storage).

31
Advanced Services Offline Messaging Contd

• Alternate model is to store the message at the
  sender-end itself.
• If the message is not delivered or the storage
  node fails, then the caller node finds the new
  storage node and records the message again
  without any user intervention.
• On boot-up a node checks for any undelivered
  message from past boot cycle, and tries to
  re-send them upon bandwidth and CPU availability.
• This model has a security flaw What if the
  sender-end is an Internet kiosk?
• To overcome this problem A third-party storage
  server can take the ownership of sending the
  message and store the message.

32
Advanced Services Multi-party Conferencing

• There are 3 methods to do a multi-party
  conference using P2P-SIP,
• A participating member can become mixer for small
  scale ad-hoc conferencing.
• Completely decentralized approach can be taken to
  exchange member information and full-mesh
  signaling is established.
• Multicast media distribution tree model can be
  used if number of senders are less at any point
  of time.
• In all these models there is a trade-off in terms
  of reliability (single point of failure in case
  of single mixer), complexity and bandwidth
  utilization.

33
Advanced Services NAT and Firewall Traversal

• In an ideal world, ISPs and corporate system
  administrators should enable their NAT and
  firewall devices with SIP proxies or application
  level gateways.
• There are two aspects to overcoming the problem
  posed by ISPs
• Automatic detection of the type of NAT and
  firewall. Detection is done at the application
  startup when the node connects to a super-node.
• Tunneling through the NAT and firewall devices
  for inbound or outbound messages. The node
  implements the Interactive Connectivity
  Establishment (ICE) algorithm for NAT traversal
  (Can use both TCP and UDP). Also every node has a
  built-in STUN (Simple Traversal of UDP through
  NAT) and TURN (Traversal Using Relay NAT) server.

34
Advanced Services Directory Service

• In any chat application people tend to search by
  first-name or last-name. Usually only partial
  names are supplied, which leads to wildcard
  character behaviour.
• Queries could also be with respect to number of
  hops from the person issuing the query.
• General DHT does not support these type of
  queries.
• These queries are supported by,
• Registering combination of first-name and
  last-name in DHT.
• Doing a blind search in acquaintances graph based
  on small number of hops-to-live.

35

Performance prediction - Scalability

• Scalability
• With N super-nodes and n nodes in the chord
  system, number of keys k per node n/N.
• REGISTER refresh rate is rs, refresh rate for
  finger table entry is rf.
• Call arrival (Possion distribution) with mean c,
  user registration (Uniform distribution) with
  mean t per user, Node churn (Possion
  distribution) with mean ?.
• Average lookup time in Chord is O(log N), as
  there are O(log N) finger table entries Node join
  leave messages generated is O( (log N)2)
• The average message rate per node is sum of the
  message rates due to refresh, call arrival, user
  registration and node join or leave
• If each node can handle C requests per second,
  then the equation C M gives the maximum
  possible number of nodes, Nmax, in the system,
  which roughly translates to Nmax 2(C/(rc))
  for large N, where r is the refresh rate and c is
  the call rate. Example If a node supports 10
  requests per second, r1min, c1per min, then max
  nodes in system 2(1030).

36

Performance prediction - Reliability

• Reliability
• What happens when a node fails? Impacted entity
  is user registration records/data in that node.
• Reliability is achieved by,
• Refresh rate is increased so that node failure
  detection happens quickly.
• User registration refresh rate is increased so
  that loss is very brief.
• User registration record is replicated at
  multiple nodes at log(N) successive nodes.

37

Performance prediction Call setup latency

• Latency
• O(log N) nodes to be looked-up before a call can
  be setup. This is a bottleneck in reducing call
  setup latency.
• In a 10000 node cluster it takes 6 hops 1-2
  seconds before call is made. This is higher than
  vanilla SIP (200ms) but can be tolerated as
  phones ring for a much higher time than 1-2
  seconds.
• P2P synchronization latency due to churn in
  nodes/records can result in multiple
  retransmissions before call setup is complete.
  Example skype takes 3-8 seconds.
• To solve this issue a hybrid model of structured
  unstructured P2P network is needed with one hop
  lookup. This solution assumes large storage space
  is available in peers.

38
Open Issues

• Security, trust and reward
• Security threats Trust, Malicious behaviour by
  nodes, DoS
• Trust Proprietary protocols like Skype has
  built-in trust among peers but open protocols
  like P2P-SIP may have malicious clients.
• Reward As in any P2P Reward nodes serving in
  DHT, punish free-riders.
• Media routing
• In case of media transfer paths we need media
  relay node in NAT and firewall scenarios.
  Stability of relay node is important as delay due
  to relay node dropping out is not tolerable.

39

Thank You !
40

Detailed slides of QoS papers
41
CS234/NetSys210 Advanced Topics in
Networking Spring 2012 Assessment of VoIP
quality over Internet backbones Athina P.
Markopoulou, Fouad A. Tobagi, Mansour J.
Karam Presentation by Swaroop Kashyap Tiptur
Srinivasa Anirudh Ramesh Iyer Tameem Anwar
42
Introduction to Voice over IP (VoIP)
43
Introduction to VoIP

• What is VoIP ?
• VoIP Protocols
• MGCP Media Gateway Control Protocol
• RVP over IP Remote Voice Protocol Over IP
  Specification
• SGCP Simple Gateway Control Protocol
• SIP Session Initiation Protocol
• Skinny Skinny Client Control Protocol (SCCP)

44
VoIP Architecture

• Coding Decoding of Analog Voice
• Analog-to-Digital and Digital-to-Analog
  conversions
• Compression
• Signaling
• Call setup tear down
• Resource coding negotiation
• Transport of Bearer Traffic
• Voice packet transmission
• Routing
• Support of quality of service
• Numbering
• Phone number, IP address

45
VoIP Architecture - Contd

• A naive representation of VoIP system as shown
  in this paper
• Sender
• Encoder Does sampling and creates a constant bit
  rate stream
• Packetizer encapsulates speech samples into
  packets of equal sizes
• Receiver
• Playback buffer An important component which
  absorbs delay/jitter
• De-packetizer Decoder Reconstruct the speech
  signal

46
VoIP Architecture - Contd

• Codecs They are used to convert an analog voice
  signal to digitally encoded version and
  vice-versa.
• Codecs vary in the sound quality, bandwidth
  required, computational requirements, etc.
• Each service, program, phone, gateway, etc
  typically supports several different codecs, and
  when talking to each other, negotiate which codec
  they will use.

47
Assessment of VoIP

• Goal of the paper Assessment of VoIP over
  today' Internet
• Organization of the paper
• Section I II describes the components of VoIP
  system under evaluation ? Already Covered in
  Architecture
• Section III presents the quality measures used
  for assessing the impairments over the network
  and our methodology for rating a call
• Section IV describes the probe measurements and
  classify the traces into categories according to
  their delay and loss characteristics
• Section V application of proposed methodology to
  the traces discuss numerical results pertaining
  to phone calls quality
• Section VI concludes the paper

48
Assessment of VoIP Quality Metrics

• Delay and Loss
• Mean Opinion Score (MOS)
• It is a numerical method of expressing voice and
  video quality
• MOS is quite subjective, as it is based figures
  that result from what is perceived by people
  during tests
• MOS is expressed in one number, from 1 to 5, 1
  being the worst and 5 the best
• Emodel
• It is a computational model, standardized by
  ITU-T, that uses transmission parameters to
  predict the subjective quality of packetized
  voice

49
Assessment of VoIP Quality Metrics Contd

• How do we obtain a subjective score (MOS) from
  metrics like delay, jitter, loss?
• Emodel provides a solution to this problem by
  defining a computation model (based on
  statistical analysis)
• First we calcluate R rating of a call using the
  formula, R (Ro - Is) - Id - Ie A, where Ro
  (effect of noise) and Is (accounting for loud
  connection and quantization) terms are intrinsic
  to the voice signal. Id and Ie capture the effect
  of delay and signal distortion respectively.
• More about Id and Ie in next slide.

50
Assessment of VoIP Quality Metrics Contd

• Delay impairment factor Id, models the quality
  degradation due to one-way or mouth-to-ear(m2e)
  delay.
• Id Idte(m2e,EL2) Idle(m2e,EL1) Idd(m2e),
  where terms Idte(m2e,EL2) and Idle(m2e,EL1)
  capture the impairments due to talker and
  listener echo respectively. EL 8 (infinite echo
  loss) corresponds to perfect echo cancellation.
  Term Idd(m2e) captures the interactivity
  impairment when the m2e delay is large, even with
  perfect echo cancellation.
• Along-with this definition a new parameter task
  is defined to measure interactive nature of call
  (eg business conference which consists of
  continuous bursts by different people OR normal
  Bob-Alice conversation). 1 is defined to be
  highly interactive (random and bursty), 6 is
  defined to be relaxed and free conversation.

51
Assessment of VoIP Quality Metrics Contd

• Loss impairment Ie captures the distortion of the
  original voice signal due to low-rate codec, and
  packet loss in both the network and the playback
  buffer.
• Table shows effect of Codec on this parameter.

52
Assessment of VoIP Quality Metrics Contd

• Observed facts about Ie and Id in different calls
  (more about test setup in actual methodology and
  results slides)

53
Assessment of VoIP Quality Metrics Contd

• Finally, now that we have R rating by evaluation
  Ie and Id we can get MOS by using the
  translation shown below. (Needed because PSTN
  call quality is measured in MOS and is gt3.6)

54
Assessment of VoIP Quality Metrics Contd

• An important factor/issue while measuring the
  parameters that we have ssen so far is the issue
  of using correct time interval to measure these
  parameters.
• For long duration calls a natural approach is to
  divide the call duration into fixed time
  intervals and assess the quality of each interval
  independently.
• Evaluating each interval in terms of Ie leads to
  transitions between plateaus of quality. (Also
  accounts human memory about bad/good transitions
  by using distribution for forgetting time
  interval when calculating instantaneous MOS. Bad
  transition is high loss aka bursts and Good
  transition is low loss aka gaps)

55
Assessment of VoIP Internet Measurements

• A model with adaptive playback delay is used in
  measurement to simulate real-VoIP implementations
  and experiment is conducted to measure the
  parameters involved in Call Quality.
• Probes are configured across 7 different Service
  Providers in 5 major cities to measure the
  parameters for a duration of 2.5 days.
• A probe packet (based on codec scheme and
  adaptive playback delay) consists of 50 Bytes of
  data sent every 10ms. 50Bytes 10Bytes of data
  40Bytes of RTP/TCP/IP headers).
• This corresponds to G.729 rate of 8Kbps
• A total of 43 paths are tested to assess their
  support for VoIP and create a model/rules to
  ensure Internet backbone can support toll-quality
  VoIP.

56
Assessment of VoIP Internet Measurements

• The following facts could be observed from the
  probe/trace data
• The paths could be classified into different
  types (A,B, C, D, E). Paths of type A and B
  connect ASH, EWR and AND on the east coast and
  have low propagation delays, i.e. below 10ms.
  Paths of type C, D on E on the other hand,
  connect cities across the US.
• Based on the variable component of delay, paths
  of type A and C have practically no queuing.
  Paths of type B and D have in general low
  queuing, except for clustered delay spikes. Paths
  of type E are coast-to-coast loaded paths. The
  queuing component is high and the delay varies
  slowly in a short time scale.

57
Assessment of VoIP Internet Measurements

• Only 3 out of the 43 paths had consistently no
  loss during the 2.5 days observed. The rest of
  them incurred loss durations that varied from
  10ms up to 33.72sec, although the average loss
  rateswere low (e.g. lt0.2).
• 6 out of 7 providers experienced outage periods
  10-220sec for 1-2 times per day. These occurred
  due to convergence delay following route change.
• 0.5-2sec loss durations were correlated with
  delay spikes.
• The number of out-of-order packets was
  negligible.

58
Assessment of VoIP Results

• So far we have covered Call Quality Metrics,
  Model to assess quality and Trace data
  obtained from probes
• To tie it all together Model obtained from Call
  Quality Metrics is applied to Trace data to
  validate the model and understand parameters that
  drive VoIP quality

59
Assessment of VoIP Results

• Evaluating the trace data using an example a 15
  minute call on path of type E , it can be seen
  that a fixed playout of 100ms ensures that Ie is
  low and optimal region is 200ms for fixed
  playout (MOS 4).
• Even adaptive model has a playout close to it
  (122ms) with max MOS 3.6

60
Assessment of VoIP Results Contd

• For the trace example in previous slide, the
  loss-delay tradeoff graph with variation in type
  of task and echo-loss looks like,

61
Assessment of VoIP Results Contd

• Looking at instantaneous quality of calls
  initiated at random times spread over an hour, it
  can be seen that 150ms playback delay is optimal
  choice as only 6 have final MOS lt3.6
• 100ms playback results in high percentage of
  calls in low MOS region.
• Adaptive playback delay shows linear nature

62
Assessment of VoIP Results Contd

• Looking at instantaneous quality of calls
  initiated at random times spread over an entire
  day, it can be seen that 150ms playback delay is
  optimal choice as only 6 have final MOS lt3.6
• 100ms playback results in high percentage of
  calls in low MOS region.
• And adaptive playback delay shows no particular
  performance benefit with 10 of calls having MOS
  lt 3.5

63
Assessment of VoIP Results Contd

• The previous graphs/results were based on time
  duration on same path.
• Impact of different types of paths are,
• Paths of type A and B have low delay and low
  delay variability. A fixed playback delay of
  100ms is sufficient for MOS gt 4
• Paths of type B and D exhibit periodically
  clusters of high spikes. Performance is low in
  both fixed adaptive delays. Only a very high
  fixed delay results in 10 of calls with MOS lt
  3.5

64
Assessment of VoIP Results Contd

• Combining inferences drawn for different times
  and paths the following results are obtained,
• Some ISP backbones (i.e. those that are
  over-provisioned and have low delay variability,
  namely types A and C) are indeed able to provide
  high quality VoIP today. This is true for both
  short and long distance paths.
• Highly loaded paths (type E) as well as some
  over-provisioned paths exhibiting frequent delay
  spikes (types B and D) have poor VoIP
  performance. Under the best scenarios these paths
  are barely able to provide acceptable (MOS gt 3.6)
  VoIP service.
• The poor VoIP performance on loaded backbones
  (type E) makes a strong case for separating voice
  and giving it priority. It also needs a more
  sophisticated handling than over-provisioning.
• There exists a tradeoff between delay and buffer
  loss. Maximum MOS is more sensitive to an
  increase in loss rather than to an increase in
  delay. The reason for this is that the underlying
  Ie curves are sharper than the Id curves.

65
Assessment of VoIP Results Contd

• Combining inferences drawn for different times
  and paths the following results are obtained,
• The need for adaptive playout comes when EITHER
  the delay is high and there is no margin for
  overestimating it OR, when the delay is unknown
  and the receiver does not know how to select an
  appropriate fixed value.
• Tuning of parameters such as thresholds for spike
  detection, weights used for moving averages
  cannot be realistically optimized (proven with
  observations).
• Delay estimation mechanism has weakness of using
  TCP-like prediction (p d4v) which tends to
  over-estimate delays beyond what is appropriate
  to preserve interactivity. Also adapting at the
  beginning of talkspurts fails to react to short
  lived spikes while it still unnecessarily leads
  to high delays.

66
Assessment of VoIP Conclusion

• After seeing the results and considering,
• Delay and loss
• Voice quality
• Playback buffer schemes
• Conclusion
• Many paths performed poorly for VoIP traffic
  due to long delays and large delay variability
• Playback buffer schemes tradeoff between data
  loss and increased delay in the buffer
• Overall rated VoIP as poor
• Mark voice traffic and give it preferential
  treatment
• Choose playback buffer scheme to match delay
  pattern

67

Questions?
68
CS234/NetSys210 Advanced Topics in
Networking Spring 2012
Improving VoIP Quality thorugh Path Switching Shu
Tao , Kuai Xu, Antonio Estepa, Teng Fei ,Lixin
Gao ,Roch Guerin, Jim Kurose, Don Towsley,
Zhi-Li Zhang
Presentation by Swaroop Kashyap Tiptur
Srinivasa Anirudh Ramesh Iyer Tameem Anwar
69

• Introduction
• i)VoIP
• ii) Factors affecting VoIP Quality
• Architecture of Path Switching System
• How Path Quality is Estimated
• Issues about Path Switching
• Performance Evaluation

70
(No Transcript)
71
VoIP requires minimum service guarantees that go
beyond the best-effort structure of todays IP
networks
72

• 1) Network Factors
• Packet Loss
• Delay Jitter
• Network Delay
• 2) Application Factors
• Playout Buffers
• Codec Performance

73
(No Transcript)
74
(No Transcript)
75
APS Gateway Components

• Path Prober
• Send UDP probes , which are generated to emulate
  the behavior of a VoIP call, on each available
  path periodically
• Receive feedbacks containing the delay and loss
  statistics.
• Path Quality Estimator
• Translate network path measurements into
  application quality estimates using the E-model

76
APS Gateway Components (Cont)

• Path Selector
• Dynamically decide the best performing path for
  each voice call
• Packet Forwarder
• Integrated with the path selector
• Forward the voice packet along the corresponding
  path.

77

• Perceived voice quality is typically measured by
  the Mean Opinion Score (MOS), a subjective
  quality score that ranges from 1 ( unacceptable)
  to 5 (excellent).
• The MOS method is from Methods for subjective
  determination of transmission quality , ITU-T
  Recommendation P.800, August 1996.

78
(No Transcript)
79
The ITU-T E-Model defines a R-factor ,that ranges
from 0 to 100 ,combines different aspects of
voice quality impairments The E-Model method
is from The E-Model , a computational model for
use in transition planning , ITU-T
Recommendation G.107, March 2003
80

• Among all of the factors in Equation 1 ,only Id
  and Ie are typically considered variable in a
  VoIP System . Using default values for all other
  factors reduces the model to
• R 94.2 Ie - Id
• When R94.2 , the value of MOS is 4.4

81

• Ie accounts for impairment caused by both
  encoding and transmission losses

82

• Id accounts for impairment caused by network
  delay, codec-related delay and play out buffering
  delay.

83
(No Transcript)
84

• Path Quality Prediction
• Estimating the Benefits of Path Switching
• Time- Scale Adaptive Path Switching Algorithm

85
(No Transcript)
86
(No Transcript)
87
A typical example that shows how path switching
avoids quality degradations (a) quality
variations on the two paths (top), and (b) the
resulting quality when path switching is applied
(bottom).
88
Conclusion

• The effectiveness and benefits of path switching
  was examined, and its feasibility was
  demonstrated with the help of a of a prototype
  application-driven path switching gateway
• With sufficient path diversity, path switching is
  indeed capable of yielding meaningful
  improvements in voice quality
• The experiments also highlighted the benefit of
  adaptive decisions, especially in light of the
  often changing nature of the time scale at which
  network congestion takes place.

89

• The study suggests that by exploiting the
  inherent path diversity of the Internet,
  application-driven path switching is a viable
  option in providing quality-of-service to
  applications
• Intelligently creating and exploiting path
  diversity via mechanisms such as overlays and
  dynamic path switching is therefore an important
  avenue to meet and improve the quality-of-service
  of applications
• There is ongoing research being done to pursue
  these issues further in the context of hybrid
  wired/wireless networks and other applications
  such as video.

90
The APS gateways
91
(No Transcript)
92
CS234/NetSys210 Advanced Topics in
Networking Spring 2012
QoS-Enabled Voice support in the Next-Generation
Internet Issues, Existing Approaches and
Challenges Bo Li, Mounir Hamdi, Dongyi Jiang,
and Xi-Ren Cao, Hong Kong University of Science,
Technology ,Y. Thomas Hou, Fujitsu Laboratories
of America
Presentation by Swaroop Kashyap Tiptur
Srinivasa Anirudh Ramesh Iyer Tameem Anwar
93
Introduction

• While the Internet has served as a research and
  education vehicle for more than two decades, the
  last few years have witnessed its tremendous
  growth and its great potential for providing a
  wide variety of services
• Over the past few years, reliability and quality
  have quickly improved, and Internet telephony is
  now one of the fastest growing industries.
• The reason behind Internet telephonys success
  is that it can potentially bring enormous
  benefits to end users, telecommunication
  companies, and carriers.

94
Advantages of IP Telephony

• It is cheaper for end users to make an Inter-
  net telephony call than a circuit-switched call,
  mainly because operators can avoid paying
  interconnect charges.
• Internet telephony gives new operators an easy
  and cost-efficient way to compete with incumbent
  operators
• Engineering economics favors Internet telephony.
  While a circuit-switched telephony call takes up
  to 64 kb/s, an Internet telephony call only takes
  up to 68 kb/s and possibly even less bandwidth.

95
Advantages of IP Telephony(Cont.)

• In the longer term, it offers exciting new
  value-added opportunities such as high fidelity
  stereo conferencing bridges, Internet multicast
  conferencing, and telephony distance learning
  applications, phone directories and screen
  popping via IP, or even voice Web browsing.
• Internet telephony gives carriers the ability to
  manage a single network handling both voice and
  data. Internet telephony will also create end
  user opportunities and demand for new services.

96
Latency

• Latency has been constantly undergoing changes
  and will continue to improve, driven by three
  factors
• Improved gateways (developers are just
  beginning to squeeze latency out of the first
  generation of products)
• Deployment over private networks by deploying
  gateways on private circuits, organizations
• Service providers can control the bandwidth
  utilization and hence latency

97
Objective

• The objective of this article is to review the
  recent developments and key enabling
  technologies in providing QoS supporting for
  voice communications in the next-generation
  Internet. The rest of the article is organized as
  follows. We first review the existing
  technologies in supporting VoIP networks,
  especially the basic mechanisms in the IETF
  Internet telephony architecture.

98
Internet Telephony Standards

• To support Internet telephony and other related
  applications, standards are being recommended
  and developed to insure interoperability
• ITU H.323 specification for Internet telephony
  is gaining widespread acceptance among software
  vendors
• The IETF is developing protocols such as Session
  Initiation Protocol (SIP) for multimedia
  session initiation, and RTSP for controlling
  multimedia servers on the Internet that can work
  together with H.323.

99
Internet Telephony Standards(cont)

• Real-Time Transport Protocol (RTP) is
  interwoven with all the above protocols. It is
  used by H.323 terminals as the transport
  protocol for multimedia both SIP and RTSP were
  designed to control multimedia sessions
  delivered over RTP.
• To this end, it guarantees that each participant
  in a session has a unique identifier, providing
  applications a way to de-multiplex packets from
  different users.
• RTP also contains a control component, called
  the Real-Time Control Protocol (RTCP). It is
  multicast to the same multicast group as RTP,
  but on a different port number.

100
Internet Telephony Standards(cont)

• One of the key components supporting VoIP is a
  signaling protocol, which has to provide the
  following functions
• User location
• Session establishment
• Session negotiation
• Call participant management
• Feature invocation
• Within the IETF, two protocols are defined
  to implement these tasks Session Initiation
  Protocol(SIP) and Session Description Protocol
  (SDP)

101
Session Initiation Protocol

• SIP is used to initiate a session between
  users. It provides user location services, call
  establishment, call participant management, and
  limited feature invocation.
• SIP is a client- server protocol. This means
  that requests are generated by one entity
  (client), and sent to a receiving entity (the
  server), which process them.
• There are three types of servers. SIP requests
  can traverse many proxy servers, each of which
  receives a request and forwards to the next-
  hop server, which may be another proxy server or
  the final user agency server. A server may also
  act as a redirect server, informing the
  client of the next-hop server so that the client
  can contact it directly.

102
Session Description Protocol

• SDP is used to describe multimedia sessions
  for both telephony and distributed applications.
  The protocol includes several kinds of
  information, as follows
• Media streams convey the type for each media
  stream. For each media stream, the destination
  address (unicast or multicast ) is indicated
  by Address.
• Ports define the UDP port numbers for each
  sending or/and receiving stream.
• Payload type conveys the media formats that can
  be used during the session.
• For a broadcast-style session such as a
  television program, start and stop times convey
  the start, stop, and repeat times of the
  session
• Originator names the originator of the session
  and how that person can be contacted.

103
Basic Mechanisms in H.323

• H.323 are a series of Recommendations of the
  ITU-T to enable multimedia communications in
  packet switched networks H.323 is designed to
  extend the traditionally circuit-based services
  including audiovisual and multimedia conferencing
  services into packet-based networks
• One of the primary objectives of H.323 is the
  interoperability with the existing
  circuit-switching systems (PSTN and ISDN).
• The basic elements defined in H.323
  architecture are terminals, gateways,
  gatekeepers, and multipoint control units (MCUs),
  in which the terminals, gateways, and MCUs are
  collectively referred as endpoints.

104

The H.323 Protocol Stack
105
Basic Mechanisms in H.323(cont)

• A gateway, as the name suggests, is an
  intermediate device to provide interoperation
  between H.323 compliant devices and non-H.323
  devices, in particular PSTN and ISDN devices.
• A gatekeeper manages a set of registered
  endpoints, collectively referred as a zone. Its
  main functions include call admission (or call
  authorization), address resolution, and other
  management-related functions.
• An MCU provides the necessary control needed
  for multiparty video conferences. It contains
  two logical components a multipoint controller
  (MC) for call control coordination and a
  multipoint processor (MP) to handle audio or
  video mixing.

106
H.323 Protocol Phases
107
The IETF Differentiated Services Framework

• The Diffserv architecture is based on a simple
  model where traffic entering a network is
  classified and possibly conditioned at the
  boundaries of the network, and assigned to
  different behavior aggregates (BAs), with each
  BA being identified by a single Diffserv code-
  point (DSCP)
• Sophisticated classification, marking, policing,
  and shaping operations need only be implemented
  at network boundaries or hosts .
• A Diffserv architecture can be specified by
  defining or implementing the following four
  components
• The services provided to a traffic aggregate
• The traffic conditioning functions and PHBs used
  to realize the services
• The Diffserv field value (DSCP) used to mark
  packets to select a PHB
• The particular node mechanism to realize a PHB.

108
There are two approaches to provide Diffserv
The first approach specifies the QoS in
deterministically or statistically quantitative
terms of throughput, delay, jitter, and/or loss.
Such approach is called quantitative Diffserv.
The second approach specifies the services in
terms of some relative priority of access to
network resources and is called priority based
Diffserv.
109
The CISCO Solution Enterprise IP Telephony

• The Cisco solution for IP telephony in
  enterprise networks includes hardware, such as
  switches, routers, IP/PSTN gateways, desktop IP
  phones, and software, such as the call manager
• By using routers and gateways to connect the
  PBX, voice traffic can be carried over data IP
  networks. Call management soft- ware and IP
  telephones are deployed in the existing IP
  networks at each remote site. This will reduce
  the cost of WAN consolidation while at the same
  time eliminating the cost of installing a second
  network at each remote location.
• Packet classification identifies and cate-
  gorizes network traffic into multiple classes.
  The Cisco IP phone can set the IPv4 ToS at the
  ingress to the network.

110
The CISCO Solution Enterprise IP Telephony

• The QoS guarantees are primarily provided by
  two mechanisms
• The call manager equipped with a resource
  reservation protocol (e.g., RSVP)
• A priority queue mechanism.
• The priority queue mechanism is maintained
  in the core routers, and is responsible for
  high-speedswitching and transport as well as
  congestion avoidance.

111
The Cisco Data and IP telephony configuration
112
Lucent Gateway Solution ForService Provider
Networks

• In this architecture an H.323 or SIP-compliant
  terminal is connected to the IP switch or
  router. The edge switches or routers serve as
  access points and concentrators for the core IP
  network, which comprises higher-capacity IP
  routers or switches.
• Two gateways are added to the IP network
  architecture as interfaces to the PSTN. The
  first added is a connection gateway (CG), which
  performs signaling interworking between the IP
  protocol and PSTN protocols. The second is a
  voice gateway (VG), which converts time division
  multiplexed signals into IP packet and vice versa

113
Lucent IP and PSTN Architecture
114
Difference between Lucent and Cisco Solutions

• The Lucent router implements a straightforward
  scheme for QoS. It simply extracts ToS
  information from incoming IP packets and sets up
  a series of prioritized queues. These queues
  can control packet flow based on the CoS value,
  which allows the router to prioritize voice
  data and move fax data to a lower priority,
  thereby minimizing delay on real-time information
  at the expense of less time-critical
  information.
• The difference between these two approaches lies
  in the fact that the Cisco system is targeted
  for the enterprise network, in which per flow
  end-to- end QoS guarantee is possible
• The Lucent approach is used for carrier
  networks, which is more scalable but relies on
  the underlying IP network to provide the needed
  QoS.

115
Conclusion

• There has been significant work done to establish
  the foundation to support VoIP. However, much
  remains to be done in order to ensure the QoS for
  VoIP and for multimedia traffic in general.
• This article surveys the existing technologies
  to support VoIP, in particular the basic
  mechanisms in the IETF Internet telephony
  architecture and ITU-T H.323-related
  recommendations.
• It then reviews the IETF QoS framework and major
  components in providing such QoS guarantees,
  including the Intserv and Diffserv models.
• In addition, this article also presents two
  leading companies (Cisco and Lucent) solutions to
  offering IP telephony services
• One another major issue currently under active
  development is internetworking with legacy net-
  works (i.e., PSTN). There are a number of
  proposals within the IEFT, in particular
  Media Gateway Control Protocol (MGCP).

116
Questions?