Buffer Sizing for 802.11 Based Networks
Abstract:-
We consider the sizing of network buffers in 802.11based networks. Wireless networks face a number of fundamentalissues that do not arise in wired networks. We demonstrate thatthe use of fixed size buffers in 802.11 networks inevitably leads toeither undesirable channel under-utilization or unnecessary highdelays. We present two novel dynamic buffer sizing algorithmsthat achieve high throughput while maintaining low delay acrossa wide range of network conditions. Experimental measurementsdemonstrate the utility of the proposed algorithms in a productionWLAN and a lab test bed.
Architecture:-
Algorithm Details:-
- eBDP algorithm
- Dynamic buffer sizing algorithm,
- Adaptive Limit Tuning (ALT) Feedback Algorithm
Explanation:-
The algorithms in thispaper perform similarly when the DCF is used and when TCPACKs are prioritized using the EDCA as in. Per flowbehavior does, of course, differ due to the inherent unfairnessin the DCF and we therefore mainly present results using theEDCA to avoid flow-level unfairness
Abbreviation
DCF-Distributed Coordinated Function (DCF)
EDCA -Enhanced distributed channel access (EDCA)
Existing System:-
The distribution of packet servicetimes is also strongly dependent on the WLAN offered load.This directly affects the burstiness of transmissions and sobuffering requirements.
IEEE 802.11b support up to 11 MBps, sometimes this is not enough – far lower than 100 Mbps fast Ethernet
In co-existing environment, the probability of frequency collision for one 802.11 frame vary from 48% ~62%
Disadvantages
Unaware of interference from/to other networks
Weak security policy
Poor performance (coverage, throughput, capacity, security)
Unstable service
Customer dissatisfaction
Proposed System:-
In this paper we demonstrate the majorperformance costs associated with the use of fixed buffer sizesin 802.11WLANs and present two novel dynamicbuffer sizing algorithms that achievesignificant performance gains. The stability of the feedbackloop induced by the adaptation is analyzed, including whencascaded with the feedback loop created by TCP congestioncontrol action.using the A* algorithm proposed in thispaper, the RTTs observed when repeating the same experimentfall to only 90-130 ms. This reduction in delay does not comeat the cost of reduced throughput, i.e., the measured throughputwith the A* algorithm and the default buffers is similar
in this paper is on TCP trafficsince this continues to constitute the bulk of traffic in modernnetworks (80–90% of current Internet traffic and also ofWLAN traffic ), although we extend consideration to UDPtraffic at various points during the discussion and also duringour experimental tests.
Advantages
The reduction in network delay not only benefits UDPtraffic, but also short-lived TCP connections
Comes from easy maintenance, cabling cost, working efficiency and accuracy
Network can be established in a new location just by moving the PCs!
Main Modules:-
1. Buffer Sizing
Buffers play a key role in 802.11/802.11e wireless networks.To illustrate this, we present measurements from theproduction WLAN of the Hamilton Institute, which show thatthe current state of the art which makes use of fixed sizebuffers, can easily lead to poor performance.. We recorded RTTsbefore and after one wireless station started to download a37MByte file from a web-site. Before starting the download,we pinged the access point (AP) from a laptop 5 times, eachtime sending 100 ping packets. The RTTs reported by theping program was between 2.6-3.2 ms. However, after startingThis work is supported by Irish Research Council for Science, Engineeringand Technology and Science Foundation Ireland Grant 07/IN.1/I901.the download and allowing it to continue for a while (tolet the congestion control algorithm of TCP probe for theavailable bandwidth), the RTTs to the AP hugely increasedto 2900-3400 ms. During the test, normal services such asweb browsing experienced obvious pauses/lags on wirelessstations using the network. Closer inspection revealed that thebuffer occupancy at the AP exceeded 200 packets most of thetime and reached 250 packets from time to time during thetest. Note that the increase in measured RTT could be almostentirely attributed to the resulting queuing delay at the AP, andindicates that a more sophisticated approach to buffer sizingis required.
2. IEEE 802.11 Media Access Control (MAC)
Measured distribution of the MAClayer service time when there are 2 and 12 stations active. Itcan be seen that the mean service time changes by over anorder of magnitude as the number of stations varies. Observealso from these measured distributions that there are significantfluctuations in the service time for a given fixed load. This is adirect consequence of the stochastic nature of the CSMA/CAcontention mechanism used by the 802.11/802.11e MAC.
3. TCP/IP packet in 802.11
Consider a WLAN consisting of n client stations eachcarrying one TCP upload flow. The TCP ACKs are transmittedby the wireless AP. In this case TCP ACK packets can beeasily queued/dropped due to the fact that the basic 802.11DCF ensures that stations win a roughly equal number oftransmission opportunities. Namely, while the data packetsfor the n flows have an aggregate n/(n + 1) share of thetransmission opportunities the TCP ACKs for the n flows haveonly a 1/(n+1) share. Issues of this sort are known to lead tosignificant unfairness amongst TCP flows but can be readilyresolved using 802.11e functionality by treating TCP ACKs asa separate traffic class which is assigned higher priority. With regard to throughput efficiency, the algorithms in thispaper perform similarly when the DCF is used and when TCPACKs are prioritized using the EDCA as in. Per flowbehavior does, of course, differ due to the inherent unfairnessin the DCF and we therefore mainly present results using theEDCA to avoid flow-level unfairness.
4. IEEE802.11e Simulation
in this paper is on TCP trafficsince this continues to constitute the bulk of traffic in modernnetworks (80–90% of current Internet traffic and also of WLAN traffic), although we extend consideration to UDPtraffic at various points during the discussion and also duringour experimental tests.
Fig:WLAN topology used in simulations. Wired link bandwidth100Mbps.
Compared to sizing buffers in wired routers, a number offundamental new issues arise when considering 802.11-basednetworks. Firstly, unlike wired networks, wireless transmissionsare inherently broadcast in nature which leads to thepacket service times at different stations in a WLAN beingstrongly coupled. For example, the basic 802.11 DCF ensuresthat the wireless stations in a WLAN win a roughly equalnumber of transmission opportunities, hence, the meanpacket service time at a station is an order of magnitude longerwhen 10 other stations are active than when only a singlestation is active. Consequently, the buffering requirements ateach station would also differ, depending on the number ofother active stations in the WLAN
In this paper, in addition to extensive simulationresults we also present experimental measurements demonstratingthe utility of the proposedalgorithms in a test bedlocated in office environment and with realistic traffic. Thislatter includes a mix of TCP and UDP traffic
5. Traffic Mix,Adaptive Limit Tuning (ALT)
We configure the traffic mix on the network to capturethe complexity of real networks in order to help gain greaterconfidence in the practical utility of the proposed buffer sizingapproach.
Fig. shows example time histories of the buffer sizeand occupancy at the AP with a fixed buffer size of 400packets and when the A* algorithm is used for dynamic buffersizing. Note that in this example the 400 packet buffer nevercompletely fills. Instead the buffer occupancy has a peak valueof around 250 packets. This is due to non-congestive packetlosses caused by channel noise which prevent the TCP congestion window from growing tocompletely fill the buffer. Nevertheless, it can be seen that thebuffer rarely empties and thus it is sufficient to provide anindication of the throughput when the wireless link is fullyutilized.
System Requirements:
Hardware Requirement:
Minimum 1.1 GHz PROCESSOR should be on the computer.
128 MB RAM.
20 GB HDD.
1.44 MB FDD.
52x CD-ROM Drive.
MONITORS at 800x600 minimum resolution at 256 colors minimum.
I/O, One or two button mouse and standard 101-key keyboard.
Software Requirement:
Operating System : Windows 95/98/2000/NT4.0.
Technology : JAVA, JFC(Swing), JMF
Development IDE : Eclipse 3.x