Scaling CSMA/CD to 1000 Mb/s
Simulation Results
Mohan V. Kalkunte
Ian Crayford

Advanced Micro Devices
I/O and Networks Division
12-March-1996

Introduction

o    Performance characterization by simulation models

o    Performance characterized in terms of:
     o    Packet transmission rate
     o    Network Utilization
     o    Data throughput utilization
     o    Collision Frequency
     o    Collision Likelihood
     o    excessiveColliisionError frequency
     o    Access Latency

     o    Packet transmission rate
          Number of packets transmitted per second

     o    Network Utilization
          The percentage of time the network is busy (good packets + header)

     o    Data througput utilization
          The ratio of data throughput to network bandwidth (good packets only)

     o    Collision Frequency
          The number of collisions per second 

     o    Collision Likelihood
          The probability of one or more collisions per successfully 
          transmitted packet

     o    excessiveCollisionError frequency
          The number of packets/sec > attemptLimit

     o    access latency 
          the waiting time of a frame when it is at the head of the MAC 
          queue until successful transmission


Setup

o       4 Node network
o       15 Node Network
o       100% offered load per station
o       1 Repeater hub
o       Traffic flows 0-1, 1-2, 2-3, 3-0

Setup

o       The following link lengths were simulated
     o       2 m
     o       100 m
     o       400 m
     o       600 m
     o       800 m
     o       1000 m

o       The following parameters were adjusted

     o   MinFrameSize

     o   slotTime

o       The parameters were stepped through the following values

     o   64 bytes, 128 bytes, 174 bytes, 256 bytes, 512 bytes

o       The following data rates were used

     o   100 Mbps, 1000Mbps

Setup

o       Four streams of UDP frames are generated independently and 
        simultaneously

o       Simulation Model was run for 10 seconds for each case

o       Statistics collected were aggregated over all stations


o       Model Parameters:

     o  Fiber delay 0.505 ns/m (one way)

     o  Repeater Hub delay 46 bit times (one way)

     o  IPG 96 bit times

     o  1500 byte data packets

Results

o       Key to interpret tables
o       trxpps = aggregate transmit/receive packets per second
o       ntwkut = network utilization
o       dtut = data throughput utilization
o       colps = aggregate collisions per second
o       collhood = collision likelihood
o       oerrps = aggregate excessiveCollisionErrors per second
o       aclate_m = mean access latency (microseconds)
o       acclate_s = standard error of access latency (microseconds)

Results

Table: Performance Measures for constant topology, 100Mbps network

diameter (m)       4           4           4           4           4
slottime (bytes)  64         128         174         256         512
trxpps (pkt/s)  8002.7      8020        8024        8030.8      8036.0
ntwkut             0.9669      0.9778      0.9796      0.9805      0.9802
dtut               0.9603      0.9622      0.9630      0.9638      0.9635
colps           4604.0      2466.9      1813.4      1248.8       644.8
collhood           0.2585      0.1434      0.1059      0.0742      0.0495
oerrps (pkt/s)   150.1        79.1        57.6        39.4        19.8
aclate_m (ms)  10.97,9.67,  6.06,3.57,  8.44,4.54,  6.00,8.85,  13.11,2.53,
               11.51,8.39   8.90,4.69   9.35,11.80  5.54,5.66   7.89,10.54
aclate_s (ms)   2.52,2.30,  2.43,0.29,  3.08,1.24   3.18,3.75,  9.50,0.125,
                2.64,2.06   3.61,0.89   2.86,4.39   7.30,1.72   4.53,7.67

Results

Utilization Vs. Slot time (Constant toplogy)

[Diagram of utilization as a function of slot time]

Results

Table: Performance Measures for varying topology &slot time, 100 Mbps ntwk

diameter (m)     200         800        1200        1600        2000
slottime (bytes)  64         128         174         256         512
trxpps (pkt/s)  7964        7752        7624.7      7513        7421
ntwkut             0.9720      0.9463      0.9309      0.9170      0.9058
dtut               0.9556      0.9302      0.9151      0.9012      0.8904
colps           4661        2391.6      1806.2      1288         639
collhood           0.261       0.143       0.110       0.0772      0.0418
oerrps (pkt/s)   153          76.8        57.9        39.5        20.3
aclate_m (ms) 8.52, 7.32,  14.22, 9.12, 16.12,12.92, 15.52,15.72, 13.12,22.12,
              7.92,10.72    8.12,11.32  19.22,12.22  17.62,11.52  12.62,13.32
aclate_s (ms) 1.75,1.47,    3.15,1.32    4.04, 2.33,  4.12, 4.18,  0.32, 6.66,
              1.58,2.41     4.94,2.25    5.81, 2.06   4.53, 0.32   0.10, 0.42

Results

Utilization Vs. Slot time/Network Diameter

[Diagram of utilization vs slot time for maximum diameter]

Results

o       Data rate 1000 Mbps
o       Slot time 512 bytes
o       100 m fiber links

Table: Performance Measures at 1000 Mbps

no. of stations      4            15
diameter (m)       200           200
slottime (bytes)   512           512
trxpps (pkt/s)   74230         71082
ntwkut               0.9062        0.8676
dtut                 0.8908        0.8532
colps             6324         29965
collhood             0.0406        0.1850
oerrps (pkt/s)     200           866
aclate_m (ms)   1.992, 1.792, 19.492, 16.792,
                1.592, 1.592  18.392, 19.992
aclate_s (ms)   0.22,  0.16,   2.11,   2.04, 
                0.14,  0.11    2.09, 2.43

Conclusions

o       Simulation Model Results match more closely with 
        experimental setup for longer fiber lengths

o       No significant degradation of throughput for large number of 
        stations