presente convocatoria y diferentes financiadores:
12.3. Tipos de evaluación.
Measurements were made to determine the through put that the adapter can sustain for received and transmitted frames. It is important to understand how throughput is related to the load, the bursti ness of frame arrivals, the percent XM I interference, and the frame size. This section presents the results of the throughput measurements as functions of these parameters.
Received Throughput as a Function of the Load The graph shown in Figure 3 is the resul t of several experiments conducted by varying the load for 33-byte received frames. The frame arrival rates depend on the load and the arrival rate distribution. As mentioned earlier, the model is capable of simu lating traffic with different arrival patterns. Figure 3 shows that, with an exponential arrival pattern, the throughput increases at a rate proportional to the
Digital Technical journal Vol. 3 No. 3 Summer 1991
Performance Analysis of a High-speed FDDI Adapter
1 00 0 80
f- z
::J o o._ U I w 60 � c:Q ::J (/J o t:: a: c:J 40 I <{ f- � w 5 20 0 20 40 60 80 1 00 1 20 KEY: LOAD (MEGABITS/SECOND) EXPONENTIAL CONSTANTFigure 3 Receive Throughput as a Function
of the Load for a 33-byte Frame
load up to a certain point, and then gradually decreases until the load is 100 Mb/s. The decrease in throughput is caused by the Joss of resources due to excessive loading.
We simulated traffic with a constant arrival pat tern and conducted the same experiments. These results are also shown in Figure 3. Observe that the point of maximum throughput and the rate at which the throughput decreases after reaching the maximum vary with the arrival pattern of traffic. After performing experiments on other frame sizes, we concluded that there is no fixed relationship between the maximum achievable throughput and the throughput at FDDI saturation (i.e ., 100-Mb/s load). Also, there is graceful degradation in through put after the peak.
Receive Throughput for Four- and Five-mode Workloads We measured adapter receive through put for four- and five-mode workloads with a load of 100 Mb/s. The XMI interference workload was var ied, and the results are presented in Figure 4. The adapter can receive the workload at 100 Mb/s, if the
XMI interference workload remains moderate. Figure 4 also shows that there is very l ittle differ ence in performance between the four- and five mode workloads. Large frames constitute a major part of both workloads, and larger frames can be eas ily supported by DEMFA at full FDDI data bandwidth.
Receive Throughput as a Function of Frame Size
Figure 5 shows the throughput as a function of the frame size and the XMI interference workload, with
DEMFA attached to an XMI (CPU) bus. Smal ler frames
Network Performance and Adapters 1 20 20 0 KEY: 20 40 60
XMI INTERFERENCE (PERCENT)
D---0 FOUR-MODE WORKLOAD
+-• • • -+ FIVE-MODE WORKLOAD
Figure 4 Receive Throughput as a Function of XMI!nterference for an
XMI (CPU) Bus Configuration
80
have a lower throughput rate than larger ones because of high controi/data overhead. Since con trol transactions consume bandwidth, the band width available for data movement is reduced . Consequently, the overal l throughput rate is lower. Another reason for lower adapter throughput is the
XMI u til ization by traffic from other nodes on the
XMI bus. This XMI interference results in less avail able XMI bandwidth for the adapter and hence, less throughput. 1 40 1 20 - - · - 20 20 50 200 500 2000 5000 1 0 1 00 1 000 1 0000
FRAME SIZE (BYTES) KEY:
XMI INTERFERENCE WORKLOAD 60 PERCENT
72
40 PE RCENT 20 PE RCENT
0 PE RCENT
Figure 5 Receive Throughput as a Function of the Frame Size for an
XMI (CPU) Bus Configuration
The adapter throughpu t for an XM.I (110) bus configuration d iffers only slightly from that for an
XM I (CPU) bus configuration. Any differences that exist are for frames smal ler than 64 bytes, since the adapter experiences a per-frame latency cost because the memory is not local to the XMI bus. Transmit Throughput for Four- and Five-mode Workloads Figure 6 il lustrates the transmit throughput for a four-mode workload as a function of the XMI interference. We performed simulations to obtain throughput data for the DEMFA when attached to an XMI (CPU) bus or to an XMI (1/0) bus. Throughput for the XMI (CPU) bus configuration is 100 Mb/s and is insensitive to low, XMI interf<:rcnce loads. Whereas, Xl\11 (110) bus configuration mea surements are negatively affected by all levels of
XMI interference traffic. The higher read latency that is inherent to an XMI (i/O) bus configuration degrades further with increasing interference traf fic. In addition the degradation appears to be lin ear. The throughputs observed for the five-mode work loads are very simi lar to the data shown in Figure 6.
Transmit Throughput as a Function of the Frame Size Figure 7 shows the throughput as a function of the frame size when the DEMFA is attached to an
X M I (CPU) bus. Throughput is also presented for
various XMI interference workloads. As in the case of receive throughput, transmit throughput degrades as the frame size decreases and the X M I
interference load increases. This degradation is 1 20 20 0 KEY: . .. 20 40 60 X M I INTERFERENCE (PERCENT)
D---0 XMI (CPU) BUS +-· · · ·+ XMI ( I/O) BUS
80
Figure 6 Transmit Throughput as a Function of X.MI Interference for a
Four-mode Workload
140
1 20
20
20 50 200 500 2000 5000
1 0 1 00 1 000 1 0000
FRAME SIZE (BYTES) KEY:
XMI INTERFERENCE WORKLOAD 60 PERCENT 40 PERCENT
20 PERCENT 0 PERCENT
FiguTe 7 Transmit Throughput as a Function of the Frame Size for an
Xll1I (CPU) Bus Configuration
again attributed to high control/data overhead and lower Xt'\11 bandwidth availabil ity.
Figure 8 shows adapter transmit throughput as a fu nction of the frame size for an Xt\11 1 (110) bus con figuration. The transmit throughput is less when the OEM FA is used with an XMI (1/0) bus rather than with an XMI (CPU) bus, due to the larger amount of
1 40
.. - - - -
20
50 200 500 2000 5000
1 0 1 0 0 1 000 1 0000
FRAME SIZE (BYTES) KEY:
XMI INTERFERENCE WORKLOAD 60 PERCENT 40 P E RCENT 20 P E RCENT
0 P E RCENT
Figure 8 Transmit Throughput as a Function of the Frame Size for the
XMJ (flO) Bus Configuration
Digital Technical journal Vol. 3 No. 3 Summer 1991
Performance Analysis of a High-speed FDDJ Adapter
read access time resulting from the XMI (110) bus configuration. The transmit operation consists mainly of read transactions and hence, this latency is crucial to transmit performance.