1. Prestented by Zhi-Sheng, Lin
Revenue Optimized IPTV Admission Control Using Empirical Effective Bandwidth Estimation
IEEE TRANSACTIONS ON BROADCASTING, VOL. 54, NO. 3, SEPTEMBER 2008
Alan Davy, Member, IEEE, Dmitri Botvich, Member, IEEE,
and Brendan Jennings, Member, IEEE
Prestented by Zhi-Sheng, Lin

2. Outline What Problem with Streaming Video?
Two Approaches to Solve This Problem
Proposed Admission Control Framework
Other Admission Control
Simulation Model and Experimental Results
Contribution of This Paper
Some Drawback in This Paper

3. What Problems with Streaming Video?
A problem with streaming video in congestion network
Packet are dropped randomly.
Packet loss & delay increased.
Incur bad QoS.
Two approaches to solve this problem. (next slide)

4. Some Approaches to Solve This Problem
First
Selectively adapt the quality of a subset of streams. (ex: adjust the transmission rate)
Second
Use admission control (AC) techniques.
Seek to ensure that the adequate bandwidth is available within the network to sustain the high QoS targets for every flow that is accepted.
AC can be divided into two types. (next slide)

5. Two Types of Admission Control
First: Parameter based admission control. (PBAC)
Pirori knowledge exists for the bandwidth requirements of each traffic source. (ex: peak rate, mean rate)
Second: Measurement based admission control.
Makes decision based on measurements taken in real time from the network. (ex: mean throughput and variance of traffic aggregates)
Hybrid of both:
Experience Based AC (EBAC).
Measurement Based and a priori Traffic Descriptor AC (MTAC).
PBAC 缺點: 那些資訊不會永遠被建立
MBAC缺點: 不正確的測量會導致不正確的有效頻寬估計

6. Proposed Admission Control Framework
Scope:
Unicast video-on-demand (VoD).
QoS:
Targets on packet delay.
This framework is a hybrid AC:
Analyze a trace relating to an aggregate of traffic flows collected over a set time interval.
Descriptor: peak throughputs of the requesting flows.  priori information
透過量測某一段時間的effective bandwidth的值”R”
R+近來的Traffic bandwidth peak throughput用來當作新的衡量標準
如果第二步驟的值可以滿足QoS目標，則接收該串流

7. Proposed admission control framework
This framework consists of:
A. Empirical Estimation of Effective Bandwidth
B. Empirical Admission Control (EAC) Algorithm
C. Revenue Maximizing Empirical Admission
Control (RMEAC) Algorithm

8. A. Empirical Estimation of Effective Bandwidth
Effective bandwidth can be defined for different types of QoS targets. (ex: delay, loss, or both)
This paper addresses delay only.
For example: delay target = (50ms, 0.001), that is only 0.1% of traffic is allowed to be delayed by more than 50ms.

10. A. Empirical Estimation of Effective Bandwidth
2019/4/29
A. Empirical Estimation of Effective Bandwidth
Delay targets = {delaymax, pdelay}
For instance, {50ms, 0.01}
// Current queue volume
// 封包可以處理完的時間點
To estimate the effective bandwidth of a particular traffic source on the network, we take a recorded packet trace of that source.
Initialization
// 累計的處理封包量
// 累計的延遲量
// 針對R速度與delaymax
// 計算最大的queue容量

11. A. Empirical Estimation of Effective Bandwidth
Processing the packets in the trace to calculate the delay that over the maximum delay.

12. A. Empirical Estimation of Effective Bandwidth
2019/4/29
A. Empirical Estimation of Effective Bandwidth
// 允許的誤差值
The effective bandwidth lies between these two values.
在reference [23]有提到為什麼effective bandwidth要介於mean rate跟peak rate之間

13. A. Empirical Estimation of Effective Bandwidth
2019/4/29
A. Empirical Estimation of Effective Bandwidth
Wow!!
Binary search here.
// 找出來的queue service rate
// 用來當effective bandwidth.

14. A. Empirical Estimation of Effective Bandwidth
Example of function calcViolations(delaymax, R, TM)

15. B. Empirical Admission Control (EAC) Algorithm

16. B. Empirical Admission Control (EAC) Algorithm
Empirical concept
Descriptor

17. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm
The objective is to use historical information about content item request arrivals, together with associated cost and resource requirements, to maximize expected revenue.

18. Item i 產生的收入 Item i 的持續時間 在 t-t’ 到 t 這段時間 要求item i 的次數 在 t 到 t+t’ 這段時間
marginal utility
Marginal臨界的

19. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm
Marginal utility
The probability of an arrival of an
additional request for the item
during the interval.

20. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm

21. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm

22. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm

23. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm

24. C. Revenue Maximizing Empirical Admission Control (RMEAC) Algorithm

25. Other Admission Control
Experience Based AC (EBAC).
Measurement Based and a priori Traffic Descriptor AC (MTAC).
Parameter Based AC (PBAC).

26. Simulation Model Service provider
When bandwidth used up to 90% of the capacity,
start admission control.
2019/4/29

27. Simulation Model (1) User Profile
2019/4/29
Simulation Model
(1) User Profile
Use a standard Poisson arrival process to generate arrivals with the expected mean arrival rate calculated from the corresponding hour period within Fig.2.
2019/4/29

28. Simulation Model (2) Service Popularity and the Pareto Distribution
2019/4/29
Simulation Model
(2) Service Popularity and the Pareto Distribution
Most popular content will contribute to a high percentage of the overall requests. (80:20 rule)
(3) Traffic Models and Characteristics
Video frame trace with MPEG-4 and H.263 format.
Simplifying assumption: customer will view the item for complete duration.
80:20 rule.
2019/4/29

29. Simulation Model i: item rank (class1-3)
req.arr.rate: request arrival rate
R(i): revenue, cost
p(i): peak rate
Ti: duration
2019/4/29

30. Simulation Model (4) Pricing Model (5) QoS Violation Measurement
TABLE IV  class 1 to class 3, the popular content is more expensive.
(5) QoS Violation Measurement
Measure QoS violations within the network by collecting a packet trace at a point following admission control and processing it through our FIFO queue algorithm.
Simulation Result (see next Slide)
2019/4/29

31. Reflection of statistical multiplexing effect
2019/4/29
Reflection of statistical multiplexing effect
EAC
More accurately predict the level of bandwidth necessary to ensure QoS targets on traffic are maintained.
As traffic is multiplexed, the required effective bandwidth for the aggregated traffic is reduced in proportion to the overall mean throughput of service flows admitted.
Effective bandwidth coefficient = prediction of bandwidth required by each AC algorithm to perform AC 除於 current mean throughput
Effective bandwidth coefficient = prediction of bandwidth required by each AC algorithm to perform AC / current mean throughput
2019/4/29

32. Utilization QoS target on packet delay of (0.02s, 0.0001) MTAC EAC
2019/4/29

33. QoS violation
QoS target on packet delay of (0.02s, )
2019/4/29

34. Table: comparison of different AC algorithm in this scenario
2019/4/29
Table: comparison of different AC algorithm in this scenario
QoS target on packet delay of (0.02s, )
AC algorithm
Utilization of bandwidth
OoS violations
else
EAC (this paper)
~50 Mbps
None
An appropriate control of bandwidth with respect to QoS targets.
PBAC
~25 Mbps
Under-utilization of available resources.
MTAC
~75 Mbps
High
Under-estimation of required resources.
EBAC
~85 Mbps
A little in this case
Over-estimates the maximum link utilization.
2019/4/29

35. With relaxed QoS targets -- utilization
QoS target on packet delay of (0.04s, )
2019/4/29

36. With relaxed QoS targets – Qos violation
QoS target on packet delay of (0.04s, )
2019/4/29

37. Analysis of RMEAC Versus EAC – class 1 item
2019/4/29
Analysis of RMEAC Versus EAC – class 1 item
RMEAC: prioritize higher value services.
EAC: First-come-first-served
2019/4/29

38. Analysis of RMEAC Versus EAC – class 2 item
2019/4/29

39. Analysis of RMEAC Versus EAC – class 3 item
2019/4/29

40. Contribution of This Paper
reaching the appropriate trade-off between ensuring QoS targets are met and not over-allocating bandwidth.
RMEAC can outperform EAC in terms of maximizing the revenue generated for the service provider by admitted traffic flows.

41. Some Drawback in This Paper
Mentioned in this paper
Central vs Distributed management
Some simplifying assumption
Ex: admitted flows are not interrupted by the user.

42. Some Drawback in This Paper
要列表格來說參數代表意義，就請詳細列出來，而不是有些在文章內提及
演算一有一列縮排錯誤
演算法三第一步驟最後兩列最好也縮排一下，不然就重複計算了
在606頁右下，Pricing Model提到Class 1是最多要求的，但從表二來看卻不然
圖三有錯誤EEAC應為EAC

43. ~Thanks for your listening~