congestion control in ndn

27
Congestion Control in NDN SHAHNEEL SIDDIQUI

Upload: shahneel-siddiqui

Post on 21-Apr-2017

114 views

Category:

Engineering


4 download

TRANSCRIPT

Page 1: Congestion Control in NDN

Congestion Control in NDN

SHAHNEEL SIDDIQUI

Page 2: Congestion Control in NDN

Presentation Outline▪ Brief Introduction on NDN

▪ NDN Performance Analysis

▪ Overload Characteristics of NDN Network

▪ Performance Deterioration Mechanism

▪ Congestion Control Methods▪ Method 1: Caching Control▪ Method 2: Interest Forwarding Control

▪ Evaluation of Caching Control

▪ Evaluation of Forwarding Control

▪ Conclusions

2/28

Page 3: Congestion Control in NDN

Brief Introduction on NDN

▪ NDN uses two types of packets: Interest & Data

▪ Each NDN router maintains three kind of tables▪ Content Store (CS),

▪ Pending Interest Table (PIT)

▪ Forwarding Information Base (FIB).

▪ The CS caches received Data Packet temporarily

▪ The PIT records an Interest Packet that has been forwarded, waiting for the Data Packet to return

▪ The FIB corresponds to the routing table in IP network except that it contains data name instead of IP address.

3/28

Page 4: Congestion Control in NDN

NDN Performance Analysis

▪ The simulations are ran by changing the number of users from 1600 to 2000

▪ Each user generates the content request with the same rate, so the number of content requests is proportional to the number of users

▪ NDN is expected to achieve more efficient content delivery owing to in-network caching than IP network

▪ However, the router cache size is limited. It is assumed that every router has the cache capacity that is 5% of the total amount of contents

4/28

Page 5: Congestion Control in NDN

Overload Characteristics of NDN Network

▪ The figure below shows the average value of the maximum content acquisition time for each user as well as the number of request retransmission

▪ It can be seen that both of the content acquisition time and the number of re-request are extended with increase in the number of users

▪ The timeout period for the user to retransmit the Interest was assumed to 5 seconds.

5/28

Page 6: Congestion Control in NDN

Overload Characteristics of NDN Network (Continued)

▪ To explain the simulation results, the notations are as shown below

▪ The router RA and router RB are connected by a bidirectional link.

▪ The link from RA to RB is LA,B and its usage rate is ρ (LA, B)

▪ Similarly, the link from RB to RA is LB,A and its usage rate is ρ (LB, A)

▪ If the utilization of a given link exceeds the threshold level (say, 80%), the link is referred to a congestion link

6/28

Page 7: Congestion Control in NDN

Overload Characteristics of NDN Network (Continued)

▪ For link utilization, when the number of users was 2000, the congestion links were L34,0, L35,2, L36,2 and L2,4

▪ The time variation of the congested link utilization is shown in Figure

▪ As the number of users increases, the link utilization also increases.

7/28

Page 8: Congestion Control in NDN

Overload Characteristics of NDN Network (Continued)

▪ Particularly, utilization of L2,4 remained at almost 100% during the simulation period.

▪ Due to the limited router cache capacity, content that does not fit in the cache will be cached in the upstream router along the path.

▪ As a result, the utilization of the upstream link gradually rises.

▪ Link congestion is considered to be generated in this way.

8/28

Page 9: Congestion Control in NDN

Overload Characteristics of NDN Network (Continued)

▪ The router whose incoming link is congested (e.g., Router RA with congestion link LB,A)

▪ Routers corresponding to this condition were R0, R2 and R4

▪ For router RA the cache hit rate of interest packets towards the link LA,B was measured

▪ It can be seen that the cache hit rate is reduced as the number of user increases

9/28

Page 10: Congestion Control in NDN

Performance Deterioration Mechanism

▪ It can be assumed that the performance deterioration of NDN occurs in the following mechanism as shown1. When so many content requests occur, Interests may concentrate on specific routers

within a network.

2. At the router RD where Interests are concentrated, a large amount of Data corresponding to the Interests are returned and cached. When the cache is full, the cache replacement occurs frequently and cache hit rate decreases.

10/28

Page 11: Congestion Control in NDN

Performance Deterioration Mechanism (Continued)

▪ It can be assumed that the performance deterioration of NDN occurs in the following mechanism as shown

3. By a decrease in cache hit rate, many Interests are forwarded to the outgoing link LD,U

4. In response to these Interests, a large amount of Data flows into the incoming link LU,D

5. When the link LU, D becomes congested, Data Packets are dropped.

11/28

Page 12: Congestion Control in NDN

Performance Deterioration Mechanism (Continued)

▪ It can be assumed that the performance deterioration of NDN occurs in the following mechanism as shown6. Then timeout occurs on the user side, and the user retransmits the Interest to the

network.

7. The user repeats re-request until the acquisition of the desired data.

12/28

Page 13: Congestion Control in NDN

Congestion Control Methods

▪ Method 1: Caching Control

▪ In the case when the link LU,D is congested, if Data from LU,D can be cached in higher priority at the router RD, the future Interests would not be forwarded to LD,U

▪ This improves that the cache hit rate of the Interests towards LD,U and mitigates the congestion of link LD,U

▪ Method 2: Interest forwarding control▪ When the router RD detects congestion at the incoming link LU,D, it redirects the Interest to

other interface than that of LD,U.

▪ By doing so, the congestion of link LU,D would be mitigated

13/28

Page 14: Congestion Control in NDN

Method 1: Caching Control

▪ The objective of this method is to improve the cache hit rate of the Interest toward the interface whose incoming link is congested

▪ For this purpose, we modify the LRU (least recently used) data, that is the default caching policy of NDN

▪ In the case where the router RD has received a Data Packet and the cache of router RD is full, then

1. When the incoming link utilization ρ (LU, D) is less than the threshold (80%), the router RD applies the normal LRU.

2. When the incoming link utilization ρ (LU, D) is not less than the threshold value (80%), the router RD searches for the LRU data and examines whether it has been received from the congestion link LU, D. If so, the router RD searches for the next LRU data.

14/28

Page 15: Congestion Control in NDN

Method 2: Interest Forwarding Control

▪ The objective of this method is to reduce the usage rate of the congestion links

▪ For this purpose, each router detects congestion in its incoming links and if it detects congestion, it does not forward the Interest to that interface and searches for alternate paths

▪ The alternate path should be short and loop-free. For this purpose, Method 2 uses the number of hops to the Publisher at each router

15/28

Page 16: Congestion Control in NDN

Method 2: Interest Forwarding Control (Continued)

▪ In Method 2, the path (or interface) selection for interest forwarding is achieved in two phases, namely 1. Candidate Interface Selection

2. Forwarding Interface Selection

17/28

Page 17: Congestion Control in NDN

Candidate Interface Selection Phase

▪ The router selects candidate interface(s) for Interest forwarding based on the following conditions1. The interface whose incoming link utilization is less than the threshold (80%).

2. The interface having the minimum hop to the publisher

▪ In the following example, since the F1 is excluded by the condition 1, the candidate interface becomes the F2 and F3

18/28

Page 18: Congestion Control in NDN

Forwarding Interface Selection Phase

▪ Depending on the number of candidate interface N, the forwarding interface is selected as follows.▪ The interface having the minimum hop to the publisher among the candidate

interfaces.

▪ If there are more than two interfaces having the minimum hop, the interface whose incoming link utilization is the lowest is selected

19/28

Page 19: Congestion Control in NDN

EVALUATION OF PROPOSED CONTROLS

20/28

Page 20: Congestion Control in NDN

▪ First, with no control, the link utilization in the network is measured.

▪ Congestion appeared at links L35,2, L36,2, and L2,4 whose locations in the network are shown

Evaluation of Caching Control

21/28

Page 21: Congestion Control in NDN

▪ Maximum Content Acquisition Time▪ By control, the content acquisition time was reduced by 20.74% in maximum (trial 4) and 16.1%

in average

Evaluation of Caching Control (Continued)

22/28

Page 22: Congestion Control in NDN

▪ Congestion Link Utilization▪ We can see that our control mitigates the congestion of L36,2 and L35,2 by 9.3% and 7.2%

respectively▪ Meanwhile in the link L2,4, the utilization had reached 100% even under control.

Evaluation of Caching Control (Continued)

23/28

Page 23: Congestion Control in NDN

▪ Cache Hit Rate▪ In the router R4, significant improvement in cache hit rate can be seen.

▪ But opposite effect was observed in R2

▪ Because the R2 was adjacent to two congestion links (L35,2 and L36,2),

Evaluation of Caching Control (Continued)

24/28

Page 24: Congestion Control in NDN

Evaluation of Forwarding Control

▪ Maximum Content Acquisition Time▪ By control, the content acquisition time was reduced by 30.1% in maximum (trial 4) and 24.2% in

average

25/28

Page 25: Congestion Control in NDN

▪ Congestion Link Utilization▪ Without control the utilization of the congestion links has remained around 80%.▪ With control, it was improved by from 94.6% to 85.3%.

Evaluation of Forwarding Control (Continued)

26/28

Page 26: Congestion Control in NDN

Conclusions

▪ By these controls discussed earlier, link congestion was mitigated and content acquisition time was significantly improved

▪ Caching control helps mitigates link congestion indirectly by caching the data packet coming from congested link thus decreasing the interest packets that are forwarded to the outgoing link of the interface

▪ Forwarding control contributes more directly in decreasing the interest packets by redirecting them to other interfaces.

▪ Forwarding Control found to be more effective because caching control was only effective when the router is connected to a single congestion link

27/28

Page 27: Congestion Control in NDN

Thank You

Questions and Suggestions are welcomed !