INTERWORKING 2000 Bergen , 36 October

1
INTERWORKING 2000 Bergen , 3-6 October

• Design of a multi-layer bandwidth broker
  architecture
• G. Politis, P. Sampatakos, I. Venieris
• National Technical University of Athens

2
Overview (1)

• Internet - QoS
• existing technologies
• Integrated Services (IntServ)
• Differentiated Services (DiffServ)
• Bandwidth Broker
• Multi-layer Bandwidth Broker architecture

3
Overview (2)

• Resource Control Layer (RCL)
• Resource Control Point (RCP)
• Resource Control Agent (RCA)
• Application Middleware (AMW)
• Structure of RCP
• Design model
• Resource (re-)distribution
• Conclusions

4
Internet

• Internet has become part of our every-day life
• More demanding multimedia applications
• Concentrates wide area of services
• excess growth of the number of users
• insufficient mechanisms for delivering the
  traffic with the appropriate Quality of Service
  (QoS)
• enhanced mechanisms have to be deployed

5
Quality of Service

• QoS IP QoS enables a network to deliver a
  traffic flow end to end with the guaranteed
  maximum delay and guaranteed rate required by the
  user process, within the agreed error boundaries
  Roberts Lawrence,Quality IP
• QoS is not solved merely by increasing the
  capacity of the network links because merging
  points will always lead to congestion

6
Quality of Service

• QoS needs three components
• traffic handling (classification of data packets
  into separate flows, scheduling and buffer
  management algorithms)
• signaling ( allows the end-user to signal
  specific flow requirements and enables end to end
  coordination)
• provision and configuration (which network device
  performs which specific traffic handling
  mechanism. Monitoring, measurement and traffic
  engineering mechanisms for QoS guarantees
  evaluation, fixing congested links.

7
Existing Technologies

• Integrated Services (IntServ)
• Differentiated Services (DiffServ)
• Bandwidth broker

8
Existing technologies (2)

• IntServenhancement of the existing IP router
  with tasks executed in switch-based networks
  making the Internet connection oriented.
  Policing, admission control and QoS management by
  all the RSVP routers on a per IP-flow basis.
• Scalability and performance issues
• DiffServ defines a set of traffic classes to
  serve applications with similar QoS demands. Each
  Traffic Class describes the Per Hop Behavior
  (PHB) that the packets should receive in each
  network node. (PHB priority, maximum delay, drop
  probability,e.a.)

9
Existing technologies (3)

• Separates the operations of the border and core
  routers. (Border routers perform more complex
  tasks while core routers only perform
  prioritizing routing without keeping any IP flow
  information).
• But the signaling interface between end user and
  border router is missing
• Bandwidth Broker (BB)takes advantage of both
  models and provides the three key aspects
  (signalling, traffic handling, management)

10
Bandwidth Broker framework

• Assumes the separation into Administrative
  domains (or ISPs) with DiffServ capable core
  network
• the traffic is policed, shaped and marked at the
  ingress point by the Edge Devices (EDs).
• Adjacent ISPs have contracts (Service Level
  Agreement, SLA) to specify the traffic
  characteristics.
• The BB shapes the aggregated traffic according
  to the SLAs, monitors and controls the available
  bandwidth, check the availability of resources
  along the path in order to accept a reservation
  request.
• The end-user may signal his QoS requirements to
  the BB via a mechanism (IntServ/RSVP,COPS, e.a.)

11
Multi-layer BB architecture

• Assumptions
• data plane consists of DiffServ-aware routers
• single domain limitation
• compromise between hard QoS quarentees and
  complex design or sacrificing a percentage of
  network utilisation and have a simplify design
• Target network services
• delay and jitter sensitive with small IP packets,
  (Voice over IP)
• delay sensitive and high bandwidth requirements
  (Video conf.)
• packet loss sensitive, security
  requirements,small duration
• no need for quarantees

12
Resource Control Layer

• Resource Control Point (RCP)
• managing and distributing the resources
• initially configure the network resources
• Resource Control Agent (RCA)
• performs policy-based admission control
• controls an Edge Device or a Border Router
• configures the traffic conditioning parameters
• handles the users reservation requests
• Application MiddleWare (AMW)
• provides an interface that enables the end-user
  to signal his QoS requirements to the QoS
  infrastructure

13
RCL Architecture

• Each RCA is associated to a single ED
• The RCA performs local admission control
• The RCP should allocate to the RCA a resource
  share that represents the resources of the nearby
  network
• The load from the signalling processing of the
  request is distributed to the RCAs while the RCP
  only redistributes resources if needed

14
RCP Hierarchical structure

• The key is to distribute the resources to the
  RCAs according to the QoS traffic that can be
  handled by the core network
• Network hierarchical structure
• Sub-networks (backbone, regional areas,sub-areas)
  and RCAs
• not fully messed network

15
RCP - Hierarchical structure

• From the network structure the RCP structure is
  produced
• knowledge about network topology
• routing information
• customers SLAs
• The RCPs represent a set of physical links
  topologically related
• A link is member of a only one RCP of the same
  level
• The sub-areas should not be directly linked

16
RCP - Coarse Design Model
17
RCP - Coarse Design model

• RCPComponent, RCPComposite, RCPLeaf according to
  the composite pattern
• Network Description, Resources Description are
  used to create and initialise the RCPs. (by
  network database)
• RCPResources Resources assigned to the RCP
• max_bandwidth the maximum amount of bandwidth
  that could ever be assigned to the specific RCP.
  (restricted mainly by the capacity of the
  physical link)
• total_bandwidth the amount of bandwidth that is
  actually assigned to the RCP.
• spent_bandwidth the amount of bandwidth
  currently distributed to the children (in case of
  Leaf is used for the reservation requests)

18
RCP - Resource (re-)distribution

• Static approach is inefficient, a more adaptive
  mechanism should be defined
• watermarks mechanism
• two watermarks high and low watermark
• Algorithm steps (upon a reservation request)
• the requested bandwidth and the spent_bandwidth
  are accumulated
• if the result exceeds the high watermark then a
  request for more resources is made to the parent
• the parent by using the resource distribution
  algorithm will decide the amount of bandwidth
  that will distribute
• that amount is depended on the free resources of
  the parent RCP

19
Resource (re-)distribution algorithm

• if (3 req 25 of the available resources)
• give 3req
• elseif (2 req 25 of the available
  resources)
• give 2req
• elseif (req available bandwidth)
• give req
• else (request_resources from parent)
• Upon a release request
• the released bandwidth is subtracted from the
  spent_bandwidth and if the result is crossing the
  low watermark the release_resources of the parent
  is invoked
• the amount of released bandwidth is determined by
  the two watermarks. After the release the
  spent_bandwidth value should be between the two
  watermarks.

20
Conclusions

• Platform independent system (JAVA, CORBA)
• no hard QoS quarentees
• the multi-layer BB architecture solves any
  scalability issue
• the hierarchical structure of the RCPs provides a
  more effective and dynamic management of the
  network resources while keeping the design simple

21
Project Details

• AQUILA Project Adaptive resource control for QoS
  Using an IP-based Layered Architecture
• http//www-st.inf.tu-dresden.de/aquila/