Springbased Resource Management for Endtoend Services in Nextgeneration Networks - PowerPoint PPT Presentation

Title: Springbased Resource Management for Endtoend Services in Nextgeneration Networks


1
Spring-based Resource Management for End-to-end
Services in Next-generation Networks

• JOURNAL OF INFORMATION SCIENCE AND ENGINEERING
• YANG-HUI CHANG, TEIN-YAW CHUNG AND YUNG-MU CHEN
• Reporter Chia-Nung Wang

2
Outline

• Spring system and definition
• Spring-based resource allocation algorithm (SBRA)
• Spring-based resource compensation algorithm
  (SRCA)
• Reactive force suppression (RFS)
• Simulation
• Conclusion

3
Spring system and definition

• For a spring system, there are there attributes.
• Regular length (LR)
• Minimum length (Lmin)
• Spring factor (x)
• As before, a force f can decides the length of
  the spring by this formula

4
Spring system and definition 2

• Defines the resource profile of entity i for a
  specific resource as
• Ri(LRi ,Lmini ,Lmaxi ,xi)

5
Spring-based resource allocation algorithm (SBRA)

• SBRA considers only two important resources
• Computing power
• Bandwidth
• There are a simple linear relationship between
  this two resource.

The number of instructions required to process a
packet.
Correlation index
The average packet size processed by ci
6
SBRA 2
allocates a single resource type first, and
checks whether the corresponding allocation of
the other resource is acceptable.
The requirements of computing and bandwidth.
Reallocates the computing quota according to
bandwidth quota.
Reallocates the bandwidth quota according to
computing quota.
7
SBRA 3
Total computing resources allocated to the
entities.

• The computing delay
• The transmission delay
• The total delay is
• The delay bound is
• The extra delay is

8
Spring-based resource compensation algorithm
(SRCA)

• A service S in this study is defined by a set of
  serially connected entities.
• When there are a extra delay , the service
  manager invoke SRCA to transfer to other
  iNode.
• Whose entities are involved in the same service
  to compensate for .

9
SRCA 2
Will suffer an extra delay
To compensate the extra delay
10
SRCA 3

• During resource re-allocation, iNodes allocate an
  extra resource to an entity to compensate for
  .
• The extra resource may affect the QoS of some
  other services and enable many new extra delay
  transferences by
• Chain Reaction Relation (CRR).
• Starvation cycle.

11
SRCA 4

• Chain Reaction Relation (CRR)
• If P1nP2nP3nPn ?Ø, there exists a potential
  chain reaction set
• When CRR occur, it will create many control
  messages because of delay compensation.
• When an iNode re-allocates its resources, it may
  cause all chain reactions in p and generate many
  Delay Transference Messages (DT_Message).

12
SRCA 5

• Starvation cycle
• There must be a service path cycle in a network.
• If the cycle exists a resource re-allocation and
  triggers chain reactions that may cause a
  starvation cycle.
• A starvation cycle not only is responsible for a
  lengthy call set up time but also triggers an
  infinite number of DT_Message.

13
Reactive force suppression (RFS)

• The study applies a novel approach called RFS to
  reduce the number of delay transferences during a
  call setup.
• RFS employs two mechanisms to suppress the delay
  transferences
• Delay budget e
• Reactive force F R

14
RFS 2
Delay bound
Average delay

• Delay budget
• If then the iNode re-allocates
  resources to cover ?ex, but this re-allocation
  yields no any new DT_Message.
• If , iNode also
  re-allocates resources to cover ?ex and makes no
  any new DT_Message. However, entity x is removed
  from ? because it has used up its entire quota.
• Finally, if
  ,then the iNode cannot completely cover ?ex.

Idle capability gt new resource requirement for x
to covering existing resource
15
RFS 3

• Reactive force F R
• Force F R is used mainly to avoid starvation
  cycles and reduce the number of DT_Messages
  generated by CRR.
• F R exploits two attributes rf and RF.
• rf denotes the number of extra delay
  transferences during the setup of a call.
• RF denotes the maximum value of rf.
• The maximum rf equals ?n in the worst case.

16
Simulation
0.96 and 0.98 are better
The improve in RFn-2 is very close to these at
RFn
17
Simulation 2
18
Simulation 3
No e
With e
RFn
RFn-2
19
Simulation 4
20
Conclusion

• In this paper, author provide an SRCA algorithm.
• In this algorithm, an extra delay suffered by an
  entity can be compensated by other entities
  belong to the same service.
• Uses a novel scheme, RFS, to avoid excessive
  control messages.
• RFS uses two mechanisms, Delay budget e and
  Reactive force FR to limit extra delay
  transferences during the setup of a call.