Fault tolerance and load balancing in QoS provisioning with multipath MPLS paths

1
Fault tolerance and load balancing in QoS
provisioning with multipath MPLS paths
Scott Seongwook Lee, and Mario Gerla, Proc of
9th-IWQOS (Germany), June 2001

• Gyu Myoung Lee
• Broadband Network Laboratory
• Tel) [phone]

2
Term Project Schedule

• Objective
• Using IP-centric control plane (e.g., GMPLS)
• To provide QoS in optical network
• optimal network utilization and load balancing
  using multipath routing
• aggregating the resources of multiple paths and
  reducing blocking probabilities
• Key technology
• Traffic engineering and QoS routing
• Multipath routing
• GMPLS control technology in IP over optical
  network
• Target
• Survey paper related with the existing multipath
  routing mechanism
• Apply multipath routing algorithm to supporting
  QoS using GMPLS control plane in IP over optical
  network

3
Presentation Schedule

• 9. July (TUE)
• Study of Multipath Routing for QoS Provisioning
• Ref) network control and management for next
  generation internet
• 25. July (THU)
• An adaptive flow-level load control scheme for
  multipath forwarding
• 30. July (TUE)
• Dynamic constrained multipath routing for MPLS
  networks
• 1. August (THU)
• A constrained multipath traffic engineering
  scheme for MPLS networks
• 7. August (WED)
• Fault tolerance and load balancing in QoS
  provisioning with multipath MPLS paths
• 8. August (THU)
• MATE Multipath Adaptive Traffic Engineering
• 14. August (WED)
• Performance analysis of burst level bandwidth
  allocation using multipath routing reservation

4
Contents

• Abstract
• Introduction
• QoS path computation algorithms
• Single path computation
• Multiple path computation
• Path management schemes
• Simulation experiments
• Conclusion

5
Abstract

• Approaches for fault tolerance and load balancing
  in QoS provisioning using multiple alternative
  paths
• Searches for maximally disjoint (i.e., minimally
  overlapped) multiple paths
• The impact of link/node failures becomes
  significantly reduced
• The use of multiple paths renders QoS services
  more robust in unreliable network conditions
• Exploits partially disjoint paths by carefully
  selecting and retaining common links in order to
  produce more options
• Offers the benefits of load balancing in normal
  operating conditions by deploying appropriate
  call allocation methods according to traffic
  characteristics
• All the computed paths must satisfy given
  multiple QoS constraints

6
Introduction - 1

• The problem
• How to provide robust QoS services in link/node
  failure prone IP networks by provisioning
  multiple paths
• The QoS support with multiple constraints
• To guarantee fault tolerant QoS paths in the face
  of network failure
• Proposal
• Extend the single path computation algorithm with
  multiple QoS constraints to a multiple path
  computation algorithm which still looks for QoS
  satisfying paths with multiple constraints
• Search for maximally disjoint multiple paths
• Minimally overlapped with each other
• Link failures have the least impact on
  established connections
• Not limited to finding fully disjoint paths

7
Introduction - 2

• Multiple paths
• Intelligently derived by preserving common links
  which should be kept in the computation to
  produce more feasible multiple paths
• Our fault tolerant solution
• Exploits those computed multiple paths with
  planned redundancy
• Brings the benefit of load balancing
• The standard IP routing protocol support
  multipath routing
• Resequencing arriving packets, complexity in
  keeping per-flow states across multiple paths,
  end-to-end jitter, impact on TCP protocol, etc
• Various path management schemes
• The most appropriate scheme can be chosen to fit
  the applications specific QoS service
  requirements and traffic characteristics

8
Introduction - 3

• The explicit routing features of MPLS
• Aggregated traffic support, relaxing the need to
  keep per-flow state at each router
• The ability to dynamically and transparently
  reconfigure the allocated calls to another paths
  without requiring explicit per flow state change
• Using multiple paths with the path pinning
  capability of MPLS
• Low latency in response to link failures with
  high robustness and load balancing
• Fast QoS provisioning by allocating additional
  calls (in the same aggregate group)

9
QoS path computation algorithms

• Finding QoS paths with multiple QoS constraints
• Step 1 Distributing network state information
• Allows every nodes in the network to capture the
  global picture of the current network status
• Link state flooding mechanism of the conventional
  OSPF
• Step 2 Computing paths with the collected
  network state information
• This paper
• Summarize the single QoS path computation
  algorithm
• Introduce the extensions with which multiple
  paths can be computed for the specific
  conditions
• fault tolerance and load balancing

10
Single path computation - 1

• The single QoS path computation algorithm with
  multiple QoS constraints
• Minimizing the hop count
• Satisfying multiple QoS constraints
• Extended to perform the multiple QoS path
  computation
• The principal purpose of our single path and
  multipath computations
• Find the shortest (i.e., min hop) path among
  those which have enough resources to satisfy
  given multiple QoS constraints, rather than the
  shortest path with the respect to another cost
  metric (e.g., maximizing available bandwidth or
  minimizing end-to-end delay)

11
Single path computation - 2

• Definition 1. QoS metric
• G(V,E), V the set of nodes, E the set of
  links
• Each link (i,j) associated with R multiple QoS
  metric
• The QoS metric q(s,j) from node s to node j can
  be computed by the concatenation of q(s,i) and
  the QoS metric q(i,j)

Concatenation function R multiple QoS metrics are
assumed to be associated with each link
12
Single path computation - 3

• Definition 2. QoS descriptor
• D(i,j) a set of multiple QoS metrics associated
  with link (i,j)
• With Dql(i,j) lth QoS metric in D(i,j), D(s,j)
  becomes
• The QoS descriptors of the nodes must be checked
  if they satisfy given QoS constraints
• If not satisfying, the nodes of the
  non-satisfying descriptors are pruned to search
  for only the constrained paths

13
Single path computation - 4

• Definition 3. Constraint verification
• Q - a set of multiple QoS constraint, the set is
  in the same format of D each QoS metric in D is
  verified with corresponding constraints
• A Boolean function fQ(D) defined to verify if D
  satisfies Q
• When fQ(D(i)) 0, node i gets pruned by the
  algorithm
• Otherwise, it is projected and its neighbors are
  further expanded until the destination d is
  reached

14
Single path computation - 5

• Definition 4 The single path computation
  algorithm

Pruning non-qualified nodes with improper QoS
metrics
Projecting qualified nodes satisfying the QoS
constraints
Expending the QoS descriptors by increasing hop
count
Note) T a temporary set of QoS descriptors P
the set which collects all the qualified QoS
descriptors of projected nodes
15
Multiple path computation - 1

• Definition 5. Path computation conditions
• Satisfying given QoS constraints
• Maximally disjoint from already computed paths
• Minimizing hop count
• Our algorithm
• Search for multiple, maximally disjoint paths
  (i.e., with the least overlap among each other)
• The failure of a link in any of the paths will
  still relaxed the disjoint path requirement
• The path computation still consists in finding
  QoS-satisfying paths

16
Multiple path computation - 2

• Definition 6. New or old paths
• The QoS descriptor with two new variables, and o
  for old, to keep track of the degree of being n
  for new disjoint
• These two variables play the most important role
  in the multiple path computation algorithm such
  that a minimally disjoint from previously
  computed paths can be found.

17
Multiple path computation - 3

• Definition 7. The multiple path computation
  algorithm

The algorithm always looks for a newer path by
investigating all qualified nodes and does not
stop just after finding the destination
Update the new variables in the QoS descriptor, n
and o
18
Multiple path computation - 4

• The multiple path algorithm illustration

19
Path management schemes - 1

• Path management schemes
• Determine how many paths are computed for each
  destination and how calls are allocated to them
• Primarily deal with path sets
• Collections of paths which are computed with the
  same constraints
• Constraint quantization
• Allows the network to limit the number of
  possible QoS constraints sets
• Forces applications to select the one that best
  fits their traffic characteristics
• The calls with same quantized constraints and
  same destination will be aggregated together and
  use the same path sets
• Shared by all subsequence calls in the same group
  without path recomputation fast QoS
  provisioning
• Call allocation flow based packet-based load
  balancing

20
Path management schemes - 2

• Non-quantized constraint Flow-based Single Path
  (NFSP)
• The simplest approach of using the single path
  computation
• The constraints are arbitrary given
• Each path set has only one path
• Path computation is carried out whenever a new
  call request comes in
• Prone to link failures due to lack of alternate
  (standby) path
• Relatively low jitter and no out-of-sequence
  packets
• With respect to load balancing, no special gain
  is obtained

21
Path management schemes - 3

• Quantized constraint Flow-based Single Path
  (QFSP)
• The same properties of NFSP in terms of the
  network performance such as the call admission
  rate, the degree of being prone to link failure,
  the load balancing, etc
• Provide the fast QoS provisioning since it comes
  with the constraint quantization and also system
  resources are saved and shared for the path
  management

22
Path management schemes - 4

• Non-quantized constraint Packet-based Multiple
  Paths (NPMP)
• Spreads packets over multiple paths which are
  maximally disjoint from each other
• Highly robust against link failures by simply
  withdrawing broken paths from the path set and
  utilizing the rest paths as soon as link failures
  are detected
• Inevitably produces higher jitter than other
  flow-based schemes since packets follow multiple
  different paths
• Efficient when applications need fine granules of
  the QoS constraints for their specific traffic
  characteristics and they are not susceptible to
  relatively high jitter

23
Path management schemes - 5

• Quantized constraint Flow-based Multiple Paths
  (QFMP)
• Quantize the QoS constraints and allocates calls
  to specific paths in their path sets
• Benefits the low jitter property with the
  flow-based call allocation and high robustness
  against link failures with multiple paths
• Allocates calls over multiple paths evenly and
  this leads to a certain type of load balancing
  with the granule of flows
• The most efficient approach if many calls for the
  same destination arrive in unreliable conditions
• Make each flow follow the same paths resulting in
  low jitter and QoS services robust with multiple
  paths with quantized constraints

24
Path management schemes - 6

• Quantized constraint Packet-based Multiple Paths
  (QPMP)
• Spreads packets over multiple paths resulting in
  high robustness against link failure and load
  balancing with the granule of packets
• Inherits inevitably the high jitter generation by
  the nature of the packet-based call allocation
• The most cost-effective approach in terms of
  requiring system resources
• The constraint quantization relieves the heavy
  use of resources and the packet-based call
  allocation simplifies the response to link
  failures without redistribution of allocated calls

25
Path management schemes - 7

• NPMP, QFMP, and QPMP
• Run the multiple path computation algorithm and
  the number of possible multiple paths is quite
  closely related to network topology
• For the sake of simplicity, define two
  conditional factors
• a hard boundary such that the number of multiple
  paths does not exceed a certain number which
  presumably produces the maximal gains yet
  minimizing the system resources
• The end-to-end delay difference d between the
  longest and the shortest paths in the path set

26
Simulation experiments

• Simulation tool SENSE (http//www.cs.ucla.edu/NR
  L/hpi)
• Compare the five path management schemes with
  respect to fault tolerance and load balancing
• Traffic model
• IP Telephony
• 160 bytes per every 20 ms with talk probability
  of 35
• The peak rate of the traffic is 64kbps and
  average rate about 23Kbps
• 6x6 grid with all the same link capacities of
  1.5Mbps and the same link propagation delay of 1
  ms
• Call requests arrive periodically at fixed rates
  and their duration is exactly 60 sec for all
  cases
• The simulation time is 600 sec for all cases

27
Call acceptance rates of the five path management
schemes over the unreliable network

• The acceptance rates depend primarily on the
  network capacity of valid paths, not on the path
  management schemes

28
Call termination rates of accepted calls due to
link failures

• The three multiple path management schemes
  outperform the two single path management schemes
  in the face of link failure

Multiple path management schemes
29
The actual number of safely completed calls
without being affected by link failures

• The three multiple path management schemes
  outperform the two single path management schemes
  in the face of link failure

Multiple path management schemes
30
Load balancing across paths

• The benefit of using the flow-based call
  allocation
• All packets follow the same paths
• Out-of-sequence packet are avoided and no extra
  jitter incurred
• Does not take full advantage of load balancing
• In the packet-based allocation
• Packet spreading causes high jitter than in the
  flow-based allocation
• Determine the overall load variance

li(t) the network load of link i monitored at
time t, L the total number of links in the
network T the entire simulation time, V the
variance of the load averaged over the simulation
time
31
Load variance over all links in the network

• All the tree flow-based allocation schemes, NFSP,
  QFSP, and QFMP, have relatively higher variance
  than the packet based ones

32
Average jitters of all the delivered packets

• The packet-based allocation schemes, NPMP and
  QPMP, have twice as much jitter than the other
  schemes

Jitter - a running average of end-to-end delay
difference of successive paths
33
Conclusion

• Examine the conventional single path computation
  for QoS support
• Extended it to compute multiple paths more
  efficiently in order to improve fault tolerance
  and load balancing
• Multiple path computation algorithm searches for
  maximally disjoint paths
• The impact of link failure is significantly
  reduced
• Links in the network are more evenly utilized by
  spreading the network load over multiple paths
• Must satisfy given multiple QoS constraints
• The source-based computation approach
• Becomes practical in conjunction with the
  packet-forwarding technology of MPLS such that
  packets are driven to follow the favorable paths
  to meet the given QoS constraints even in
  unreliable network conditions

34
Q A - Discussion

• Traffic engineering using multipath in
  GMPLS-based optical network
• Objective
• Optical resource optimization (load balancing)
• Reduce blocking probability
• Fault tolerance
• Control plane (Routing Signaling)
• Traffic aggregation and classification
  (flow-level control)
• Constrained-based explicit route setup (consider
  multipath)
• What kind of constraints ?
• Maximum hop count constraint
• Maximum path count constraint
• Node/link affinity constraint
• Path selection policy