sfi

NDP

ENTERPRISE IRELAND

NCNRC

Communications Network Research Institute

Optimizing the Throughput of WLAN Mesh Networks

Jianhua Deng
PhD Student (November 2009 - present)
Funding Agency: China Scholarship Council

Overview

WLAN mesh networks (WMNs), are dynamically self-organized and self-configured networks that are intended to deliver wireless services for a large variety of applications in personal, local, campus, and metropolitan areas in a cost effective manner. WMNs are expected to resolve the current limitations of WLANs and to significantly improve their performance in large scale outdoor deployments.

Description of Project

In WMNs, the throughput or network throughput is the average rate at which data is successfully delivered over the wireless channel. A high throughput is the most important goal of a WMN. There are two ways to improve the throughput performance .One is to use less time when transmitting every unit data between source node and destination node. The other one is to transfer more data bits within a unit transmission time.

Owing to the noisy nature of the wireless channel, errors are introduced into the received packets which mean that corrupted packets have to be re-transmitted. Re-transmitting the packet increases the time required to deliver the data and hence reduces the throughput. For any given set of wireless channel conditions, there is an optimal packet size that minimises the probability of re-transmission and maximizes the throughput.

Approach

In this research project, maximizing the WMNs throughput through packet aggregation is investigated. A new scheme called Optimal Aggregation Sized Packet Scheme (OASPS) is adopted here. The OASPS is based around two algorithms. The first algorithm is concerned with determining the Optimal Packet Size (DOPS) of WMNs under varying wireless channel conditions. The second algorithm is concerned with how to assemble the Optimal Aggregate Packet (AOAP).

In the DOPS, it is proposed to employ an Additive-Increase Multiplicative-Decrease (AIMD) technique. Under the AIMD policy is that the size of the aggregate packet size is increased from L bits to (L+ α) bits if the destination node successfully receives the packet. Otherwise, if packet transmission fails, the aggregate packet size is decreased from L bits to βL (0 < β < 1 and the value of β is dynamic) bits, which is illustrated in Figure 1.

AIMD Example
Figure 1: An example of AIMD.

In the AOAP, there are two basic principles. The first is that the greatest number of packets should be aggregated and transmitted in unit time. The second is that no packets should be lost through buffer overflow or by exceeding their lifetime while waiting in the buffer. The aggregation algorithm should adhere as closely as possible to these two principles. In this work, 4 possible algorithms for realising efficient aggregation are proposed:

In the AOAP, as shown in Figure 2, there are two packet buffers and two hash tables. One of the two packet buffers is the receiving buffer which contains all the received packets that are to be transferred onto the next node. The other buffer is the sending buffer that contains the aggregated packet which will be transmitted at the next transmission opportunity. When the node receives an ACK indicating that the aggregate packet has been successfully received, the aggregated packet in the sending buffer will be deleted. Meanwhile, all of the packets in the receiving buffer that were contained in the aggregate packet (as identified by a unique ID number) will also be deleted. A hash table is used for storing the information of packets that are currently stored in the buffer. The diagram below shows an example of using Algorithm A to assemble packets. In the hash table, the first column is the ID number of the packet in the buffer, the second column is the packet size, the third column is the waited time of the packet and the last column is the value of N. The second hash table is used to store the ID of the packets which have been aggregated. The goal of the AOAP algorithm is to aggregate the smaller sized packets from the packet buffer into larger and optimally sized aggregate packets at every network node.

AOAP scheme
Figure 2: Two buffers and two hash tables for the AOAP showing 10 packets in the receiving buffer and 4 packets (ID=5, 8, 10, 6) that are aggregated in the sending buffer.

In Figure 3, the aggregated packet contains an aggregation of the smaller sized sub-packets. The aggregated packet has a new packet header which is shared by all sub-packets to reduce the overhead and also contains some flag bits which serve as a separator between the subpackets and also help to disaggregate in the destination node.

Aggregated packet
Figure 3: The aggregated packet and subpackets,

The objective of the Optimal Aggregation Sized Packet Scheme (OASPS) scheme is to achieve the maximal throughput for every node transmission. This will also help reduce the competition for access to the wireless channel between the nodes and make the most effective use of the available transmission opportunities.

Figure 4 shows the result when the aggregation algorithm B was used to assemble packets and was simulated in ns3.The simulation scenario involves five wireless nodes located in an area of 100 m x 100 m. One node is the source node, one is the destination node and another three nodes are the relay nodes. In the ns3 simulation, Poisson traffic model was employed and four kinds packet size were simulated by each node: 125 bytes, 250 bytes, 500 bytes and 750 bytes. The simulation time was 100 seconds. The results for the capacity and delay were compared between two wireless mesh networks one with OASPS and the other one without OASPS. Here, the increase in capacity ΔCapacity and the increase in delay ΔDelay are defined as:

ΔCapacity = (C Cn)/ Cn

ΔDelay = (D Dn) / Dn

Here C is the average capacity of WLAN with OASPS, Cn is the average capacity of the WLAN without aggregation; D is the average delay of WLAN with OASPS and Dn is the average delay of the WLAN without aggregation.

AOAP scheme
Figure 4: The capacity improvement in packet aggregation scheme compared with the normal wireless mesh network in ns3.

In figure 4, it is shown that using the packet aggregation scheme can significantly improve the capacity but also incurs more delay when it compared with the wireless mesh network without aggregation. It is shown that the Δcapacity is improved more than 300% while the ΔDelay is increased by 82.57% when the packet size is 125 bytes. When using the Poisson taffic generator with an average packet size of 750 bytes, the ΔCapacity and ΔDelay are approximately zero.

ΔCapacity is improved more than 300% while the ΔDelay is increased by 82.57% when the packet size is 125 bytes. When using the Poisson taffic generator with an average packet size of 750 bytes, the ΔCapacity and ΔDelay are approximately zero.

In this simulation, the optimal packet size (Lopt) is set to 500 bytes, α equals 100 bytes and β equals 0.5. For a given Lopt target, there are more packets to be aggregated into one aggregation packet when the packet size is smaller. It means that contention is reduced and more data is transmitted in unit time which means that ΔCapacity increases with reducing packet size for the given Lopt target. On the other side, there is more time required to allow for further packets arrival in order to achieve the Lopt target when the packets size is smaller. That also means ΔDelay for every aggregation packet also increases with decreasing packet size.

It is shown that the packet aggregation scheme can significantly improve the capacity when the packets size is small in the wireless mesh network.