1 / 45

Performance Analysis of Wireless LAN Access Points

KTH. Performance Analysis of Wireless LAN Access Points. Department of Microelectronics and Information Technology (IMIT). Royal Institute of Technology (KTH), Stockholm, Sweden. Licentiate Seminar Candidate: Iyad Al Khatib First Advisor: Prof. Rassul Ayani

harper-mclaughlin
Download Presentation

Performance Analysis of Wireless LAN Access Points

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. KTH Performance Analysis of Wireless LAN Access Points Department of Microelectronics and Information Technology (IMIT) Royal Institute of Technology (KTH), Stockholm, Sweden Licentiate Seminar Candidate: Iyad Al Khatib First Advisor: Prof. Rassul Ayani Second Advisor: Prof. Gerald Q. Maguire Jr. Opponent: Prof. George Liu, Linköpings University, Sweden May 27, 2003

  2. Outline 1. Motivation 2. AP Modeling 3. Experimental Setup 4. Tests Design 5. Results (Queuing Model) 6. AP as a Data Link 7. Feedback Model for AP Throughput 8. Conclusions 9. Open Issues and Future Work

  3. 1. Motivation • With the technological progression in data communications, I see a an obvious success for fastest (high speed) networks • The increase in WLAN deployment is evident, and users show interest in getting connected without being tethered by a wire. • Currently, most WLAN systems have Ethernet backbone, i.e. communication is via access points, rather than ad-hoc networking. • PCC (www.pcc.lth.se) project indicates that the trend will be toward even more access points attached to LANs. • High proportion of services utilized by mobile users are hosted on machines attached to wired infrastructure. • To the best of our knowledge, there has been no logical model for WLAN APs.

  4. 2. AP Modeling • Seeking to model a wireless LAN access point (WLAN AP) as a system of Reference. • Hence, discussion leads to talking about • Systems • System components • Classification of systems • Modeling • Types of Models • Model Components

  5. 2. AP Modeling 2.1. Systems • Objective: define the concept of a system, which best suits the computer networking field in general and WLANs in particular. => Simplest way is to pick the word up in a dictionary • For technical objects, Webster reads: A system is “a group of devices or artificial objects or an organization forming a network especially for distributing something or serving a common purpose.” • Scientific projection (flavor) A WLAN AP system, in the sense of the interest I have, is nothing but a collection of connected objects that interact (communicate) to accomplish some task(s).

  6. 2. AP Modeling 2.1. Systems 2.1.1. System Components • All Systems share five basic components: 1) Entity: an object inside the system, e.g. customer, packet 2) Attribute: trait of a system entity, e.g. amount of goods, packet size 3) Activity: related to time, e.g. time to purchase, response time 4) State: mode/condition of part or whole, e.g. waiting, service --> Described by State parameters 5) Event: an action that may affect the system by changing the state, --> Described by Event parameters

  7. 2. AP Modeling 2.2. System Modeling • Definition. A Model is a representation of a system aimed at studying the system • A model, by definition, is a simplification of the system. • Same system can have different Models depending on the different goals. • Types of Models: Physical and Logical. • Model Components: all or some of the five system components. For WLAN APs, we take all. • Necessary to define the boundary line, which depends on the desired parameters of the study of the model.

  8. 2. AP Modeling 2.3. AP Queuing System • We seek to model the processing in a wireless LAN access point (WLAN AP), hence, a set of assumptions was made. • Logically, we isolate the AP, define the events that occur, and the parameters related to these events. • Three types of events exist: external and internal events. • two external events: arrival,departure, and noise (eliminated). • one internal event: enter-service • Whether packets arrive from the wired or wireless sides, the model considers them as arriving packets. • Similarly, packets leaving the system are considered as departing packets regardless of the medium they go out on.

  9. 2. AP Modeling 2.3. AP Queuing System Event Parameters • We view all parameters in the investigation from the point of view of the AP itself. • Hence, when a packet leaves or arrives to the system, we are interested in the arrival time and the departure time. • Consequently, we view the relationship between arrivals and departures as the response of the system. (the parameter of interest is the response time). • Since the number of packets in the system changes when a packet enters or when a packet departs from the system, then it is a discrete-event system.

  10. 2.3. AP Queuing Model Ta = Arrival time Td = Departure time Response time (R.T.) = Td - Ta Si = Service time of packet i Wi =Waiting time of packet i Si and Wi calculated using our algorithms. Si and Wi calculated using our special algorithms. 2. AP Modeling 2.3. AP Queuing System • We define two traffic flows through the AP: • Downlink: from Ethernet to WLAN • Uplink: from WLAN to Ethernet • Analyses of experimentally recorded time stamps of arrival time and departure time values showed that the AP can be modeled as a queuing system with single queue and a single server. • Ri can be easily calculated. • However, Wi and Si are logical model parameters that can not be easily measured. Hence, we designed the SSTP (Simple Service Time Producer) algorithm.

  11. 3. Experimental Setup The testbed consists of Ethernet PCs (EPCs), WLAN PCs (WPCs), WLAN AP, and Sniffer PC (SPC). • Linux-2.2.16 • MGEN 3.2 (UDP) • tcpdump • ORINOCO Client Manager • Lucent WavePoint II • Lucent AP2000 • 11Mbps • Controlled environment

  12. 4. Tests Design • Our tests consist of two main parts: single-source-to-single-sink (SS) and multiple-sources-to-multiple-sinks (MM). • The SS part is used to extract parameters of the model, and the MM part of tests is used to check for the queue management and behavior. • Each part is made up of downlink and uplink tests. • We call the experiment in a cluster an experimental test run (ETR). The ETR is the basic unit of tests. • In each ETR, we send a stream of identical UDP datagrams from the sender(s) to the receiver(s).

  13. 4. Tests Design • For each ETR, we vary the size of the packets to be sent in a stream via increasing the payload of the UDP datagram by 32 bytes. • The maximum number of bytes we used as UDP payload was 1472, because sizes beyond the MTU may result in fragmentation. • The headers and inter-frame Spaces are thoroughly calculated before sending any traffic stream, because there is a major difference between the link frames of the Ethernet and the WLAN (in this case IEEE 802.11b).

  14. DATA SIFS Destination (receiver) ACK 4. Tests Design 4.1. The Link Scenario • In IEEE802.11, there are different link scenarios for transmission (controlled by the AP). We have used the most common one, namely DSSS with a DCF MAC. • WLAN link layer interference is described by four link layer parameters. IEEE 802.11b overhead delay can be compared to that of Ethernet (9.6μs), as shown: DIFS Source (WPC or AP) time 50μs for DIFS, ~4μs preamble, MAC header and trailer in DATA time 10μs for SIFS, ~5μs link layer ACK period

  15. 4. Tests Design 4.2. SSTP Algorithm SSTP-1.3 Ri = Wi + Si = Td - Ta

  16. 5. Results (Queuing Model) • Outcomes of tests show that the assumption of single server, single FIFS queue holds true for downlink and uplink traffic tests. • For the WLAN APs we have tested, the delay on the uplink is always smaller than on the downlink.

  17. AP1 (Lucent WavePoint II) Average Service Time (microsec) AP2 (Lucent AP2000) Average Service Time (microsec) 5. Results (Queuing Model) 5.1. Directional Service-Time Delay • Uplink is faster than Downlink. • Differs with different APs (compare performance).

  18. 5. Results (Queuing Model) 5.2. Service-Time Formula • The key result is an analytic solution for the average service-time of a wireless LAN AP: Sn = So + (n-1)r s, where (1) • n = UDP_payload (in bytes)/32B = (IP_payload[in bytes] - 8B)/32B • Sn = service time (s) for a packet with IP payload of (32n + 8)B • So = service time (s) for a packet with 40B IP payload • r = incremental difference in sec. • So is calculated by averaging the service times calculated in the different experiments, and r is calculated using linear regression of service times of different packet sizes. • So and r are the characteristic parameters of the AP.

  19. 5. Results (Queuing Model) 5.3. Analysis of Directional Service-Time • Service time on the uplink is smaller than on the downlink • We introduce the notion of the Uplink Downlink Contrast: UDC(,x) = |UST(,x)-DST(,x)| where, •  is the brand of the access point, • UST(, x) is the Uplink Service-Time of a packet with payload x bytes, for AP "", and • DST(, x) is the Downlink Service-Time of a packet with payload x bytes, for AP "".

  20. 5.3. Analysis of Directional Service-Time 5.3.1. UDC of two APs • AP1 is Lucent WavePOINT-II • AP2 is Lucent AP2000

  21. 5.3. Analysis of Directional Service-Time 5.3.2. Convergent UDC Definition. A WLAN access point, , is said to have a convergent UDC if and only if: as x increases, UDC(, x) decreases, where x is the payload in bytes.

  22. Theoretical point of intersection. In this case it is around 2200B 5.3. Analysis of Directional Service-Time 5.3.2. Convergent UDC plot • If stretched theoretically, meets the horizontal axis

  23. 5.3. Analysis of Directional Service-Time 5.3.3. Divergent UDC Definition. A WLAN access point, , is said to have a divergent UDC if and only if: as x increases, UDC(, x) increases, where x is the payload in bytes.

  24. 5.3. Analysis of Directional Service-Time 5.3.3. Divergent UDC plot

  25. 5.4. Buffer Size Estimation • we are not aware of a careful analysis of buffer size in current APs • In order to estimate the buffer size, we construct a set of experiments that purposely try to cause packet loss due to the lack of buffer capacity • We designed and implemented the Buffer Size Estimator (BSE) algorithm that detects when a packet is lost and makes use of the SSTP algorithm for extracting some parameters for buffer size estimation • Results table

  26. 5.4. Buffer Size Estimation 5.4.1. BSE Algorithm

  27. 5.4. Buffer Size Estimation 5.4.2. BSE Results APa is Lucent/ORINOCO WavePOINT-II APb is ORINOCO AP500

  28. 6. AP as a Data Link • Throughput is an important QoS parameter (link throughput studied) • Inspired by the power of Packet-Pair for FIFO-Queuing Networks, I looked at the AP as a data communication link. • Using the Link model, the throughput of a WLAN AP can be well estimated. • Throughput has been a subject of debate for WLAN APs

  29. 6. AP as a Data Link • Consider the WLAN AP as a data communication Link:

  30. 6. AP as a Data Link • Packet-pair in FIFO-queuing networks makes use of a two-packet logical model. whose parameters are the difference in arrival times of two identical packets sent from the same source to the same destination. • Assumptions fit the Tests design

  31. 6. AP as a Data Link • The packet pair property states that: • t(d,1) - t(d,0) = max{[s1/b],[t(0,1) - t(0,0)]} • where, • t(d,0) and t(d,1) are the arrival times (at the destination) of the first and second packets respectively; • t(0,0) and t(0,1) are the times of transmission of the first and second packets respectively. • s1 is the size of the second packet in bits; • b is the bandwidth of the bottleneck in bits per second.

  32. 6. AP as a Data Link Mbps Mbps, for IPv4

  33. 6. AP as a Data Link Downlink max ba Downlink min ba 1480B 40B

  34. 6. AP as a Data Link Uplink max ba Uplink min ba 1480B 40B

  35. 6. AP as a Data Link

  36. 6. AP as a Data LinkResults

  37. C(bits) C R R(sec.) + E Gf - H + - 7. Feedback Model for AP Throughput Two feedback control models for the throughput: uplink feedback model and downlink feedback model

  38. 8. Conclusions • Mathematical model for WLAN AP was presented. Our experiments showed that our assumption of a single server, single FIFS queue is correct. • The time to serve a packet going from WLAN to Ethernet is less than the time to serve an identical packet but going from Ethernet to WLAN. • Using our model and analysis, we can compare performance of different brands of WLAN APs. • The key result is that when using our model and our test design, one can get an analytic solution of the average service time in terms of payload (a strictly increasing linear function of payload).

  39. 8. Conclusions Cont’d • We use the model to analyze the absolute value of difference between uplink and downlink service time values • We define the absolute value of the difference as the UDC (Uplink-Downlink Contrast) • UDC can be characterized as: convergent or divergent • The choice of a suitable WLAN AP could be made according to the UDC characteristic of the AP with respect to the application or network

  40. 8. Conclusions Cont’d • The AP is looked upon as a link with adaptive-bandwidth bounded by two limits: the upper adaptive-bandwidth bound and the lower adaptive-bandwidth bound. • The link model of the AP builds on previous results of the AP queuing-model and its service-time analytic solution and the packet-pair property of FIFO-queuing networks, which was shown to be very suitable for the link model analysis. • The bandwidth of the link model of the AP was analyzed and found to be an increasing linear fractional transformation of payload. • A major result is that when using our model and analysis, performance of different wireless LAN APs can be compared in terms of throughput (as a QoS parameter).

  41. 9. Open Issues and Future Work Open issues are: • noise in the WLAN medium had not been considered in its effect on delay. • case of multiple access points are present is also not investigated. Moreover, effects on background traffic, though it is not an issue to be considered in benchmarking, but could be interesting. • throughput is studied for the AP without discussing the effect of multiple nodes on the wireless medium, and what that could add to the losses in quality of service. • the percentage of packets that are successfully transmitted by an AP.

  42. 9. Open Issues and Future Work • Study on the protocol (IEEE 802.11), enhancements. • More detailed model for uplink, downlink, and bi-directional traffic taking in consideration more aspects of traffic. • AP throughput model in relationship to nodes available on the IEEE 802.11 link. • Stretching the work to bridges and routers. • Investigating WLAN APs that utilize link layer combination of protocols other than the two studied in this thesis, namely 10Mbps Ethernet and IEEE 802.11b. • Developing a simulator, which is a current topic of my research (work in progress). • Traffic shaping can be used to allow emulation of a given access point, based on analyzing a set of measurements of a real access point. • Running more multimedia experiments of future applications and compare them to the future 3G wireless networks from a performance point of view.

  43. 9. Open Issues and Future Work • Investigating whether a convergent-UDC AP will have a better downlink throughput than a divergent-UDC AP. • Analyzing the uplink and downlink delays and the UDC of APs to determine the time required to serve the overhead bits on both directions (the downlink and the uplink) for a given AP. • Test my programs against a Linux machine running packet forwarding code (such as the traffic shaping code). • Research on using the service time analysis to estimate (calculate) the response time. For instance, a user may be interested to know what the average response time of a given AP with a given packet stream sequence is. • Investigate the relationship between the AP brand and the percentage of packets sent with an erroneous Transport Layer Checksum, and its effects on the throughput formula of the AP, TAPL, presented in (24).

  44. ACKNOWLEDGMENT Thanks to Daimler Sweden AB for supporting us with WLAN APs, and for their interest.

  45. Thank you for your attention, ..and have a bright day!

More Related