1
On the Optimal Configuration of Metro
Ethernet for Triple PlayAndras Kern, Istvan
Moldovan, and Tibor CinklerBudapest University
of Technology and Economics, HungaryAppeared in
IEEE 2nd Conference on Next Generation Internet
Design and Engineering (NGI), pp. 334-341, April
3-5, 2006.
Nov. 19, 2007 Chen-bin Kuo (20077202) and Young
J. Won (20063292) DPNM, POSTECH Email kuo,
yjwon_at_postech.ac.kr
2
Introduction

• This paper focuses on provisioning of the
  Ethernet aggregation to support Triple Play
  services
• Whats this about?
• About QoS provision in Metro Ethernet
• Traffic representation models for triple play
  components
• Configuration method of VLANs in Metro Ethernet
  for triple play

3
Triple Play in Metro Networks

• Promising Technology Metro Ethernet
• IEEE 802.1Q QoS standard, 8 QoS classes to the
  Ethernet
• Scalability by segmenting the network into
  independent Virtual LANs (VLANs), representing a
  different broadcast domain
• 802.1D Spanning Tree Protocol (STP)
• Defines a loop-free logical forwarding topology
  over the meshed physical topology
• Rapid Spanning Tree Protocol (RSTP) decreases the
  failover time to a few seconds
• Multiple Spanning Tree Protocol (MSTP)
• The VLANs are uniquely associated to the MSTIs
  inside a region while one MSTI aggregates the
  traffic of more VLANs

4
QoS Provisioning in Metro Ethernet

• Real Time (RT), Streaming, Transactional, and
  Best Effort (BE)
• Diffserv, compatibility with TIU-T and 3GPP
• 802.1Q compliant switches
• Hypothetical assumption 10 for RT, 20 for
  Streaming, 30 for Transactional classes
• VoIP RT
• IPTV Streaming
• HIS BE
• Acronyms
• Network Management Entity (NME)
• Policy Enforcement Points (PEPs)
• Policy Decision Point (PDP)
• High Speed Internet Access (HSI)

5
Network Architecture for Triple Play

• Policy based management model
• A policy is a set of rules controlling how to
  access to and set the priorities for the use of
  resources
• The pipe for VoIP is a point-to-point
  bidirectional pipe
• The Video service uses an unidirectional
  multicast pipe
• While the HIS service is a point-to-point pipe
  with asymmetric allocated capacity

6
VoIP Service Call Level Multiplexing

• The well-known Erlang-B formula determines the
  number of simultaneous calls required to serve a
  fixed size of population with a defined blocking
  probability threshold
• To provide an acceptable bound for call blocking
  probability
• Then, the bandwidth of a VoIP TE-pipe can be
  known
• Based on the number of parallel calls
• Based on the bandwidth requirement of the assumed
  codec

7
IPTV Service Multicast in Metro Ethernet

• Manual multicast filters based on VLAN or
  Ethernet multicast address
• Service providers can configure Ethernet switches
  manually with a multicast filter
• Dynamic multicast forwardingEthernet switches
  can also listen to the IGMP Join messages used by
  receivers to query for a multicast source
• Or the use of GMRP (GARP Multicast Registration
  Protocol) for interworking with IGMP, a few
  vendors

8
HSI Service Statistical Multiplexing (1/2)

• It does not exploit the provided bandwidth
• The bitrates of typical internet traffic
  generated by home users varies significantly
• Related Works
• F. Kelly 10 presented the theoretical basics of
  statistical multiplexing aware dimensioning
  considering different traffic models
• S. Floyd 11 proposed a simple method to
  calculate the effective bandwidth for aggregation
  of independent traffic flows based on the
  Hoeffding bound
• Large number of individual flows, the ratio of
  mean and peak rates are close to zero
• Suppose the individual traffic flows are
  independent and the aggregation has Gaussian
  distribution
• Guerin 12

9
HSI Service Statistical Multiplexing (2/2)

• Guerin 12 continues, as in (1)
• A simple, effective model
• Mean bit rates (m)
• Standard deviation epsilon of the aggregation
  considered alpha
• Bandwidth overflow probability can be adjusted by
  alpha
• Deterministic multiplexing
• The allocated capacity is the sum of the peak
  rates of the individual flows

10
Proposed Configuration Method

• A method for configuration of VLANs taking into
  account the requirements of all three services
• The authors previous work focused on providing a
  formal model dealing with unicast TE pipes
• To minimize the allocation of network resources
  for a set of traffic demands and the number of
  used MSTIs
• Presented an Integer Linear Program (ILP)
  formulation with scalability problems
• Heuristics for two subtasks
• Pipe or VLAN routing and covering them with trees
• Heuristic is needed ILP model cannot deal with
  the non-linearity by the statistical multiplexing

11
Proposed Heuristic Algorithm (1/2)

• Inputs to the algorithm
• The topologies of the trees are combined from
  paths of the assigned VLANs
• QoS constraints and the demands
• The method decomposes the problem to sequential
  search of VLAN routes and assignment to MSTIs
• It adopts the Simulated Allocation (SAL)
  metaheuristic
• Allocation operator
• Randomly select one VLAN among the unassigned
  ones
• Fit the VLAN to the already defined MSTIs
• Otherwise, create a new MSTI and assigned to the
  VLAN

12
Proposed Heuristic Algorithm (2/2)

• For unicast VLANs (VoIP and HSI) a single path is
  sought from the access to the edge node
• For multicast VLANs, the tree is determined by
  calculating independent shortest paths from each
  accesses to the sole target one after the other
• Disconnection process
• It selects the tree having the least assigned
  VLANs, and remove it

13
Case Study (1/2)

• Traffic drive configuration method produces
  high throughput gain in the topologies of
  practical interest 6
• The sizes of the demands depend on the overall
  throughput (OT) and the service ratio (SR)
  parameters
• OT is responsible for the system-wide throughput
• SR defines how to distribute the traffic between
  the classes
• Assuming
• VoIP traffic is 10, video broadcast traffic is
  30, and the best effort internet is 60
• The exact size of a demand is generated randomly
  with Gaussian distribution

14
Case Study (2/2)

• Taking the variance of the traffic into account
  ratios of the mean and peak rates
• PMR b_peak / b_mean
• VoIP 1.0, Video 1.2, Internet 2.0
• Evaluation of Results
• Figure depicts the allocated capacities (or
  network load)
• Considering statistical multiplexing, 20 higher
  throughput can be achieved
• Considering statistical multiplexing influences
  the paths calculated for VLANs
• Statistical multiplexing becomes the part of the
  optimization task

15
Conclusions

• Due to cost considerations the Ethernet becomes
  the most cost effective solution for regional and
  metropolitan networks
• Presented a QoS service architecture over
  Metropolitan Ethernet networks
• Based on traffic engineering logical channels (or
  pipes)
• Main contributions
• Propose an efficient algorithm for off-line
  configuration of metro Ethernet networks for
  triple play
• Multicast VLAN based trees to provide video
  broadcast service
• The gain of statistical multiplexing

16
Questions ?