Scalability Analysis

Simulation Machine:

CPU: Pentium4 3.0 GHz Memory: 2G RAM

OS: Linux, Fedora Core 4 with modify nctuns-2.6.11-kernel

Table 5-5 One BS to one SS with coding, without application traffic Simulation Time 40sec 80sec 120sec fec Mode Time (sec) Mem (KB) Time Mem Time Mem

Table 5-5 shows the minimum simulation execution requirements, time cost and memory usage, since it only contains management messages. The execution time is in proportion to the simulation time and the memory usage is stabled. After analyzing the call graph profile reported by the “gprof” utility, over 60% execution time is taken by convolution decoding. Because the convolution code with code rate 5/6 conveys more data than code rate 2/3, it uses fewer symbols. Therefore, since the size of management messages contained is constant, QPSK 3/4 and 16QAM 3/4 may process faster than QPSK 1/2 and 16QAM 1/2.

Table 5-6 gives the execution time with greedy traffic. It simulates 40 seconds and there are 5 seconds greedy traffic from the BS to the SS. Duo to the higher data rates, the execution time is greatly growing. Almost all computations are bound on the convolution decoding. However, because the convolution decoder computes and

accesses memory bits by bits, it is matter of course to use m the convolution decoding is the major procedu

uch CPU time. Seeing that re of simulation, the execution time is depending on the decoding data (traffic) size. In the MAC layer modules, each conn 4000 packets (configurable), this queue length must acco data rates and frame duration. Each packet contains a 1428

bytes network simulator.

Therefore th

One BS to with 5 second greedy traffi

S ime second 5 se ds y

ection can buffer at most mmodate the target

IP packet and other data structure used in the NCTUns e growth of memory usage, 8MB, is reasonable.

Table 5-6 one SS c

imulation T 40 s with con greed traffic fec Mode Time (sec) Mem (KB)

In order to verify the influence of data rate, we are experimenting with different traffic and coding configuration as Table 5-7. The first case same as Table 5-6 is experimental group. The second case disables channel coding and channel errors. In the third case, we enable channel coding and channel errors and add a dummy SS into the WiMAX network but there is no application traffic sent to it. In the final case, we add a greedy UDP traffic to the dummy SS added in previous case.

Because the channel coding and channel error factor was disabled in case 2, it simulates in several seconds. However, the throughput is always high abidingly and lost in accuracy.

Table 5-7 The influence of different configuration Simulation Time 40 second, (5 second greedy traffic)

Simulation One SS, with coding, Without Two SS, two Configuration 1 greedy traffic coding Two SS

traffic

decoding data size and double execu e resulted

In case 4, there are two greedy traf ent to the two SS respectively. Since they share the same edium, the amount of data rate are fixed. There is no much extra data to decode than case 3. Therefore, the e is sim case 3. And within the BS, there are two connections which are full of packets. It consumes another 8MB memory space

ccording to the analysis as above, the execution time is in proportion to the proc

ion limitations are desc

ere is no traffic sent to the new added SS in case 3, th

and decoding data. Because double SS are in decoding, double tion time ar .

fics s m

execution tim ilar to

s.

A

essed data rate, and the memory usage is depending on the traffics and connection queue length. Since the simulation is CPU bound, we can predict the possible ranges of execution time and memory usage according to the basic cases listed in Table 5-6 and Table 5-7.

Another scalability issue, the supported scale of WiMAX network simulation module, it depends on the NCTUns network simulator. The execut

ribed as above. Since we use the IP-CS scheme in our design and the NCTUns supports class C network, each BS node can support at most 254 SS nodes.

Futu

ted t of desi implem of our In

ion lso s and d r ss ur . ve

f tio clud n the X hn ar lt u do not

d elo ca ties in t sta I c s

interesting tasks and some optional features, which are worthy to do in the future.

oS support. Currently, we

processing retransmits MAC SDU blocks that have been lost or

ification which provides more flexible and more complex allocation. Although there are many differences between 802.16d and 802.16e, we still provide a reference

Chapter 6. re Work

We illustra he main ideas gn and entation work.

addit , we a howed prove the co rectne of o work Howe r, the unc nalities in ed i WiMA tec ology e mu itudino s. We

ev p all of the pabili defined he ndard. n this hapter, we list everal

„ QoS signaling and BW request/grant for QoS service The IEEE 802.16 Standard defines four classes of Q

use a simple approach as displacement for QoS and the functionalities of QoS is needed to supplement.

„ Automatic Repeat Request (ARQ) connection ARQ, which

garbled, can be enabled or disabled on the connection basis. Enabling ARQ on connections may increase or decrease the throughput and latency of WiMAX network. Using ARQ can help to improve reliability of TCP. The effect of ARQ on the WiMAX network maybe can compare with our results.

„ IEEE 802.16e and OFDMA

IEEE Standard 802.16e is an advanced version of 802.16. It supports both fixed and mobility, and reinforces some other features such as hold off and power saving for support mobility. The 802.16e works on the OFDMA spec

implementation for it. Based on this reference im these modules to support 802.16e.

plementation, we can extend

like s Mesh mode, it is compatible with PMP mode. Due to the inconsistency

nctionality is in demand.

„ Mesh/Relay

Another possible extension of our work is the Mesh/Relay project, or called 802.16j Mobile Multi-hop Relay (MMR). It is expected to extend coverage and enhance throughput to the non-line of sight area. Since it uses relay station a

between PMP and Mesh mode, supporting this relay fu

Chapter 7. Conclusion

imulation is a convenient approach to analyze and evaluate networks. It is more periment using real machines and more accurate than mathematical mod

famo base

simu

esigns of the MAC and physical layers. The MAC layer focuses on the management control and the integration with NCTUns. The main functions of the physical layer are performing channel coding and simulating channel model. The channel coding in our implementation performs real bit-level encoding/decoding operations to provide the highest fidelity. For channel modeling, the Rayleigh fading and an empirically based path loss model is adopted.

Furthermore, the simulation results in our implementation are presented, validated, and analyzed. We also discuss the scalability issue. In our implementation, simulations are conducted with encoding/decoding procedures operating at coding block level to obtain more realistic results at the cost of slowing down the simulation speed significantly.

In conclusion, we developed a useful WiMAX PMP-mode network simulator over NCTUns. Based on our work network researchers can develop their protocol modules over our platform, obtaining more realistic results, and analyzing their results in a more efficient way.

S

convenient than ex

eling. We developed a WiMAX 802.16 network simulation system based on the us NCTUns network simulator. In our design, the simulated network supports station nodes and two kinds of subscriber station nodes.

In previous chapters, we explained what we did and how we did to support lating 802.16 in NCTUns, including the architecture of protocol modules, the d

Reference

Vinko Erceg, et al., “An Empirically Based Path Loss Model for Wireless Channels in Suburban Environments,” IEEE Journal on selected areas in communications, Vol.17, No.7, July 1999.

C. Bushue, A. Sarajedini, R. Schwartz, D. Brandlund, T. Kaitz, and D.Trinkwon.

“Channel Models for Fixed Wireless Applications.” July 2 [1]

[2] Vinko Erceg, K.S. Hari, M.Smith, D. Baum, K. Sheikh, C. Tappenden, J. Costa,

001.

[3]

Prentice Hall, 1996, p.286.

Theodore S. Rappaport, "Wireless Communications - Principles & Practice,"

[4] Marvin K. Simon, Mohamed-Slim Alouini, "Digital Communication over Fading Channels - A Unified Approach to Performance Analysis," John Wiley & Sons, 2000, p.223. [2nd Editionp.256 (8.112)]

[5] RSCODE project, http://rscode.sourceforge.net/

[6] C. Hoymann, “Analysis and Performance Evaluation of the OFDM-based

Metropolitan Area Network IEEE 802.16,” Comp. Net., vol. 49, no. 3, Oct. 2005, pp. 341–63.