QoS Routing for MPLS Networks Employing Mobile Agents By: Sergio Gonzalez-Valenzuela and Victor C. M. Leung

1
QoS Routing for MPLS Networks Employing Mobile
AgentsBy Sergio Gonzalez-Valenzuela and Victor
C. M. Leung

• Presented by Nathan Balon

2
Introduction

• The authors of the paper propose using
  DiffServ-over-MPLS to provide QoS, the problem
  with this approach is the lack of routing support
  in the current architecture.
• The authors introduce a new routing algorithm
  that provides support for establish QoS compliant
  routes by using mobile agents.
• The goal of the paper is to introduce the uses of
  mobile agents at the network layer, to provide
  QoS routing in Internet backbone.

3
Overview

• DiffServ
• MPLS
• Mobile Agents
• Wave
• Routing Algorithm
• Results

4
DiffServ

• QoS is used to provide a consistent predictable
  data delivery service.
• In the authors framework they propose that
  DiffServ is used to provide QoS.
• The reason DiffServ was chosen because it is
  highly scaleable.
• DiffServ provides a means of categorizing and
  prioritizing network traffic flow aggregates.

5
DiffServ

• DiffServ uses Differentiated Services Code Points
  that identifies a per hop behavior (PHB) to be
  applied at each node to a packet.
• Service Level Agreements (SLA) are used to define
  the policy criteria and traffic profile.
• One problem with DiffServ is it can not work by
  itself, it needs to use another protocol to
  perform forwarding.

6
MPLS

• Multi-protocol Label Switching (MPLS) is based on
  a combination of layer 3 routing with label
  switching and forwarding.
• MPLS achieves the simplicity of a layer 2 switch
  while retaining flexibility and scalability of a
  layer 3 router.
• Each packet in an MPLS network contains a label
  which is associated with a Forward Equivalence
  Class (FEC).
• FEC provide scalability. They are used group
  flows that are forwarded in a similar manner.

7
MPLS

• The benefit of MPLS is routing decisions are made
  at the edge of a network where a packet will be
  given a label based on the FEC.
• MPLS packets are then forwarded based on the
  label, which determines the destination of a
  stream.
• At each hop a new label is inserted into the
  packet replacing the old label.
• MPLS routers contain a table with the next hop
  for a label. A packet is forwarded based looking
  up a label in table. The forwarding table avoids
  having to make costly look ups in a routing table
  and keeps the size of the routing table small.

8
MPLS Header
MPLS Shim Headers (1-n)

n
1
Network Layer Header and Packet (eg. IP)
Layer 2 Header (eg. PPP, 802.3)
4 Octets
Label Stack Entry Format
TTL
Label
Exp.
S
9
The Need for a Routing Protocol

• MPLS framework requires an external routing
  protocol to determine routes and for the
  distribution of labels.
• There is also a need to support
  multipoint-to-point connections. Where a number
  of data streams can start at separate locations
  and converge on a common node or follow the same
  path and later diverge.
• Excessive routing traffic can be avoided creating
  multiple routes to handle different QoS
  requirements which are defined by SLAs.

10
Mobile Agents

• Mobile agents are autonomous software entities,
  that can move themselves from one node to another
  in a network.
• Mobile agents can achieve a collective objective
  in a cooperative manner.
• The benefits of mobile agents are
• Agents can communicate with other agents.
• If the machine that launched the agent into the
  network was removed, the agent could continue to
  operate.
• Agents can perform parallel processing.
• One problem is there is no current support for
  mobile agents in the Internet infrastructure.

11
Wave

• The Wave paradigm is used to implement the mobile
  agents.
• Each wave is defined by a string that represents
  the agents operations and variables.
• Each wave may start their execution at a given
  node and then propagate through the network.

12
Wave Program

• One of the reason for the choosing Wave to
  implement the mobile agents is the program is
  compact.
• Wave code is approximately 20 to 50 times more
  compact then many high level languages.

13
Wave

• Each node in the network needs to run a Wave
  interpreter.
• The Wave framework also provides an interfacing
  mechanism, so that waves can communicate with
  programs written in other languages.
• Waves use the information they gather when
  traversing the network to build a Knowledge
  Network (KN).
• Wave agents are able to clone themselves in a
  virus like fashion and are then propagated to
  neighboring nodes.
• Agents can carry along with them any information
  they need and set variables at nodes to share the
  information they gather.

14
Benefits of Wave

• Strong migration
• Synchronous/asynchronous navigation
• Agent collaboration
• Flexibility
• Fault tolerance
• Autonomy
• Compactness

15
QoS Routing

• The routing scheme uses two types of agents
• A set of static agents obtain availability of
  network resources.
• A second set of agents are deployed into the
  network find QoS compliant routes.

16
Static Agents

• Static agents are created to reside on every node
  in the network.
• The purpose of static agents is to monitor the
  available resources in a DiffServ switch.
• The agents gather information such as the amount
  of bandwidth, delay and jitter.
• The static agents can update the value of a
  virtual link to a neighbor in the KN.
• A benefit of this is there is no need to flood
  the network with routing tables whenever a change
  takes place.

17
Static Agents
18
Discovering QoS Compliant Routes

• The goal at this phase is to create a tree with a
  minimized cost for a specific type of service.
• Paths that do not contribute to optimization of
  resources are pruned from the tree.
• The end result is minimization of the use of
  network resources such as the number of
  communication links needed and the number of
  labels.

19
Egress and Ingress Nodes

• Egress node A differentiated service (DS)
  boundary node whose role is handling traffic as
  it leaves a DS domain.
• Ingress node A DS boundary node whose role is
  handling traffic as it enters a DS domain.

20
Route Discovery

• Route discovery consist of 2 parts.
• Finding QoS compliant routes for each ingress
  node whose final path is the root of the tree.
• Determining if any routes overlap.

21
Finding QoS Compliant Routes

• During initialization of a switch a request from
  ingress nodes is made to a common egress node.
• A colony of mobile agents are launched from all
  ingress nodes to an egress node to create a mp2p
  tree.
• The agents clone themselves and then launched
  copies on each QoS compliant link.
• When the agent arrives at an intermediate node,
  the distance traveled is recorded in a local
  variable.
• If the agent has a shorter distance than was
  previously recorded for an agent originating from
  the same node it is able to continue.
• If the agents distance is larger, the agent is
  discarded.

22
Finding QoS Compliant Routes

• A second set of agents are then launched.
• The goal of this round of agents is to determine
  all of the possible shortest paths to the root.
• Agents are allowed to continue on their path to
  the root if their recorded distance equals that
  recorded by the previous set of agents.

23
Finding Overlapping Routes

• Next, another set of agents are launched from
  different origins to find routes that minimize
  resources by finding routes that will allow data
  to be merged.
• When the phase starts each node sends an agent to
  travel along the shortest paths that were
  previously discovered.
• When an agent arrives at a node
• The agent checks a flag to see if any other agent
  had visited that node from the same origin.
• If the agent is the first to visit the node, it
  sets a flag which tells other agents that the
  node was previously visited.

24
Finding Overlapping Routes

• A final set of agents is then set out along the
  same shortest path to determine the number of
  visits at that node from distinct origins.
• The more visits that a node has the better, the
  path is then credited a higher weight.
• Every agent will record associated weights to a
  destination node based on the amount of
  overlapping.
• The result is the egress node will contain the
  records for all the shortest paths from each
  origin to itself.

25
Creating Routes
26
MPLS Switches Can Be Updated

• Using Waves interfacing feature, the IP addresses
  for nodes found on each path can passed to the
  MPLS switch.
• The MPLS switch can then assign labels and use a
  label distribution protocol.

27
Results

• One problem that was found with the routing
  algorithm is that routing traffic can take place
  in large bursts.
• The sudden increase in the number of agents can
  cause congestion problems and queuing delays.
• The authors determine that more efficient
  navigation techniques need to be developed.

28
Arrival Pattern of Mobile Agents
29
Conclusion

• An framework such as the one the authors are
  suggested is unlikely to be embraced at the
  present time.
• To use mobile agents in core of the Internet
  infrastructure, new routers would need to be put
  in place to the support mobile agents.
• The paper was written to show longer term
  possible routing solutions.