Authors:

1
Comparative Evaluation of Adaptive Price-based
Admission Control Algorithms for Bandwidth
Allocation

• Authors
• Ch. Bouras, K. Stamos
• Research Academic Computer Technology Institute,
  Greece
• University of Patras, Computer Engineering and
  Informatics Department, Greece
• http//ru6.cti.gr

2
The DiffServ architecture

• One of the basic architectures that have been
  proposed for QoS provision in the Internet
• The Internet consists of numerous network domains
  
• One scalable entity is proposed to administrate
  resources and provide end-to-end QoS across
  network domains

3
Definition of BB

• A Bandwidth Broker is an entity responsible for
  providing QoS within a network domain. The
  Bandwidth Broker manages the resources within the
  specific domain by controlling the network load
  and by accepting or rejecting bandwidth requests
• inter-domain interface
• intra-domain interface
• routing table interface
• user/application interface
• policy manager interface
• network management interface

4
Admission control

• Has to guarantee
• fairness
• network utilization
• Types of incoming requests
• Immediate
• Book-ahead (or advance)
• Our request model
• r(tstart,tend,b,w)

5
Request states
6
Reservation requests example
7
VPN hose model (Duffield et al.)

• Bandwidth management is simplified by assigning a
  limit at the bandwidth that each edge router is
  allowed to accept in the domain
• assumes that proper dimensioning of the network
  has taken place in order to be able to support
  the hoses definition
• part of the available bandwidth for the links has
  been assigned to the management of the Bandwidth
  Broker for the DiffServ service
• the network design will guarantee that there will
  be enough resources in order to service the
  reservation requests that have been accepted
• offers the flexibility to the edge device of
  sending traffic to a set of endpoints without
  having to specify the detailed traffic matrix
• can also reduce the required size of access links
  through multiplexing gains because of the
  aggregation of flows between endpoints.

8
Adaptive admission control module

• Request queue
• Rr1, r2,, rm? Wq
• we want to find a subset Rc?R such that
  S(ri?Rc)bi ? B at any time point
• B is the total available bandwidth for the
  service
• try to maximize S(ri?Rc)bi throughout the period
  from the earliest tstart to the latest tend in
  the R set
• Rc is the set of requests that will be accepted
  by the algorithm, while requests in the set R-Rc
  will be rejected

9
Adaptive admission control module

• NP-complete problem
• approximation algorithm
• adaptive parameter for the size of R, which will
  increase if the number of requests in Wq
  increases or if the algorithm was very
  time-consuming, and decrease otherwise

10
Characteristics of the algorithm

• The algorithm is not generally optimal on network
  utilization
• trade-off in order to reduce the computation
  overhead for the BB module
• A very fast processing module (or conversely a
  low rate of admission requests) lead the
  algorithm to quickly converge to the best
  approximation of the optimal solution

11
Characteristics of the algorithm

• Rsize converges to the size of Wq as quickly as
  (1-a)t converges to near-zero values
• quite rapidly, especially if a has been chosen
  close to 1, which means a very high adaptation
  capability

12
Taking rejected requests into account
13
Client resubmissions

• wait for indication from central module
• progressively increasing time intervals between
  resubmissions
• avoid scaling problems
• prioritize resubmissions
• place in front of queue

14
Admission module enhancement

• In order for the BB to utilize resubmitted
  requests, it needs to keep a list L of the
  standby requests
• It will actively prioritize such requests in
  expense of newly received requests
• The prioritization will depend on the duration
  that a specific user has been waiting and
  resubmitting a request.

15
Resubmission interval

• A request that has been resubmitted n times, this
  means that the total time tw that has elapsed
  since the initial request will approximately be
• tw tresubmit n2/2
• if we assume Cavgltlttresubmit for a reasonably
  computationally capable admission control module
• the waiting time is increased quite rapidly as
  the number of times that the request is rejected
  increases
• This avoids constant resubmission of rejected
  requests in situations were the resources that
  are available to be allocated are significantly
  below the requested resources

16
Simulations

• higher level simulated system for evaluating the
  algorithms
• ns-2 network simulator on an Intel-based Linux PC
  with 288 MB of main RAM, Pentium III 700MHz
  Coppermine with 256 KB cache memory for
  evaluating packet-level behaviour
• parameters randomly produced within suitable
  boundaries (all requests competed for the same
  resource limit at the ingress point of the
  simulated network)

17
Simulation results Admission control adaptation
Mild adaptation
Aggressive adaptation
18
Acceptance rate
19
Network utilization
20
Average response time
21
Conclusions

• Our proposed algorithm improves on the common
  admission control modules for management of
  domain resources
• offers better utilization of the network
  resources
• keeps a balance between simplicity and
  functionality
• automatically falls back to a simpler model if
  necessary
• resubmitted requests do not obstruct the rest of
  the requests
• each resubmission significantly increases the
  possibility of acceptance
• The admission control module tries to prevent
  unnecessary resubmissions by aggressively
  discouraging end users from resubmitting requests
  if it notices that the rejection rate becomes
  exceedingly high

22
Conclusions

• the architecture so that it can be configured
  according to the specific parameters of each
  deployment
• processing power
• acceptable notification delays
• the proposed solution is suitable for a variety
  of real-world cases

23
Conclusions

• There are emerging types of networks, (e.g.
  wireless)
• strong demand for quality of service
• sophisticated solutions are applicable because of
• smaller capacity links
• frequent and more unpredictable entry of new
  users for existing resources
• this work has been taken over in order to be
  applied to wireless environments

24
Future Work

• Evaluate our model on an actual environment
• Measure our implementations resilience to
  various kinds of failures
• Securing the module operation from compromised
  components, disobedient clients, stolen/altered
  messages while transmitted on the network
• Extend model to management of resources at lower
  layers

25
Thank you

• Email Info
• Christos Bouras (bouras_at_cti.gr)
• Kostas Stamos (stamos_at_cti.gr)