Network Working Group S. Bradner
Request for Comments: 1944 Harvard University
Category: Informational J. McQuaid
Bay Networks
May 1996
Benchmarking Methodology for Network Interconnect Devices
Status of This Memo
This memo provides information for the Internet community. This memo
does not specify an Internet standard of any kind. Distribution of
this memo is unlimited.
Abstract
This document discusses and defines a number of tests that may be
used to describe the performance characteristics of a network
interconnecting device. In addition to defining the tests this
document also describes specific formats for reporting the results of
the tests. Appendix A lists the tests and conditions that we believe
should be included for specific cases and gives additional
information about testing practices. Appendix B is a reference
listing of maximum frame rates to be used with specific frame sizes
on various media and Appendix C gives some examples of frame formats
to be used in testing.
1. Introduction
Vendors often engage in "specsmanship" in an attempt to give their
products a better position in the marketplace. This often involves
"smoke & mirrors" to confuse the potential users of the products.
This document defines a specific set of tests that vendors can use to
measure and report the performance characteristics of network
devices. The results of these tests will provide the user comparable
data from different vendors with which to evaluate these devices.
A previous document, "Benchmarking Terminology for Network
Interconnect Devices" (RFC 1242), defined many of the terms that are
used in this document. The terminology document should be consulted
before attempting to make use of this document.
Bradner & McQuaid Informational [Page 1]RFC 1944 Benchmarking Methodology May 1996
2. Real world
In producing this document the authors attempted to keep in mind the
requirement that apparatus to perform the described tests must
actually be built. We do not know of "off the shelf" equipment
available to implement all of the tests but it is our opinion that
such equipment can be constructed.
3. Tests to be run
There are a number of tests described in this document. Not all of
the tests apply to all types of devices under test (DUTs). Vendors
should perform all of the tests that can be supported by a specific
type of product. The authors understand that it will take a
considerable period of time to perform all of the recommended tests
nder all of the recommended conditions. We believe that the results
are worth the effort. Appendix A lists some of the tests and
conditions that we believe should be included for specific cases.
4. Evaluating the results
Performing all of the recommended tests will result in a great deal
of data. Much of this data will not apply to the evaluation of the
devices under each circumstance. For example, the rate at which a
router forwards IPX frames will be of little use in selecting a
router for an environment that does not (and will not) support that
protocol. Evaluating even that data which is relevant to a
particular network installation will require experience which may not
be readily available. Furthermore, selection of the tests to be run
and evaluation of the test data must be done with an understanding of
generally accepted testing practices regarding repeatability,
variance and statistical significance of small numbers of trials.
5. Requirements
In this document, the words that are used to define the significance
of each particular requirement are capitalized. These words are:
* "MUST" This word, or the words "REQUIRED" and "SHALL" mean that
the item is an absolute requirement of the specification.
* "SHOULD" This word or the adjective "RECOMMENDED" means that there
may exist valid reasons in particular circumstances to ignore this
item, but the full implications should be understood and the case
carefully weighed before choosing a different course.
* "MAY" This word or the adjective "OPTIONAL" means that this item
is truly optional. One vendor may choose to include the item because
Bradner & McQuaid Informational [Page 2]RFC 1944 Benchmarking Methodology May 1996
a particular marketplace requires it or because it enhances the
product, for example; another vendor may omit the same item.
An implementation is not compliant if it fails to satisfy one or more
of the MUST requirements for the protocols it implements. An
implementation that satisfies all the MUST and all the SHOULD
requirements for its protocols is said to be "unconditionally
compliant"; one that satisfies all the MUST requirements but not all
the SHOULD requirements for its protocols is said to be
"conditionally compliant".
6. Test set up
The ideal way to implement this series of tests is to use a tester
with both transmitting and receiving ports. Connections are made
from the sending ports of the tester to the receiving ports of the
DUT and from the sending ports of the DUT back to the tester. (see
Figure 1) Since the tester both sends the test traffic and receives
it back, after the traffic has been forwarded but the DUT, the tester
can easily determine if all of the transmitted packets were received
and verify that the correct packets were received. The same
functionality can be obtained with separate transmitting and
receiving devices (see Figure 2) but unless they are remotely
controlled by some computer in a way that simulates the single
tester, the labor required to accurately perform some of the tests
(particularly the throughput test) can be prohibitive.
+------------+
| |
+------------| tester |<-------------+
| | | |
| +------------+ |
| |
| +------------+ |
| | | |
+----------->| DUT |--------------+
| |
+------------+
Figure 1
+--------+ +------------+ +----------+
| | | | | |
| sender |-------->| DUT |--------->| receiver |
| | | | | |
+--------+ +------------+ +----------+
Figure 2
Bradner & McQuaid Informational [Page 3]RFC 1944 Benchmarking Methodology May 1996
6.1 Test set up for multiple media types
Two different setups could be used to test a DUT which is used in
real-world networks to connect networks of differing media type,
local Ethernet to a backbone FDDI ring for example. The tester could
support both media types in which case the set up shown in Figure 1
would be used.
Two identical DUTs are used in the other test set up. (see Figure 3)
In many cases this set up may more accurately simulate the real
world. For example, connecting two LANs together with a WAN link or
high speed backbone. This set up would not be as good at simulating
a system where clients on a Ethernet LAN were interacting with a
server on an FDDI backbone.
+-----------+
| |
+---------------------| tester |<---------------------+
| | | |
| +-----------+ |
| |
| +----------+ +----------+ |
| | | | | |
+------->| DUT 1 |-------------->| DUT 2 |---------+
| | | |
+----------+ +----------+
Figure 3
7. DUT set up
Before starting to perform the tests, the DUT to be tested MUST be
configured following the instructions provided to the user.
Specifically, it is expected that all of the supported protocols will
be configured and enabled during this set up (See Appendix A). It is
expected that all of the tests will be run without changing the
configuration or setup of the DUT in any way other than that required
to do the specific test. For example, it is not acceptable to change
the size of frame handling buffers between tests of frame handling
rates or to disable all but one transport protocol when testing the
throughput of that protocol. It is necessary to modify the
configuration when starting a test to determine the effect of filters
on throughput, but the only change MUST be to enable the specific
filter. The DUT set up SHOULD include the normally recommended
routing update intervals and keep alive frequency. The specific
version of the software and the exact DUT configuration, including
what functions are disabled, used during the tests MUST be included
as part of the report of the results.
Bradner & McQuaid Informational [Page 4]RFC 1944 Benchmarking Methodology May 1996
8. Frame formats
The formats of the test frames to use for TCP/IP over Ethernet are
shown in Appendix C: Test Frame Formats. These exact frame formats
SHOULD be used in the tests described in this document for this
protocol/media combination and that these frames will be used as a
template for testing other protocol/media combinations. The specific
formats that are used to define the test frames for a particular test
series MUST be included in the report of the results.
9. Frame sizes
All of the described tests SHOULD be performed at a number of frame
sizes. Specifically, the sizes SHOULD include the maximum and minimum
legitimate sizes for the protocol under test on the media under test
and enough sizes in between to be able to get a full characterization
of the DUT performance. Except where noted, at least five frame
sizes SHOULD be tested for each test condition.
Theoretically the minimum size UDP Echo request frame would consist
of an IP header (minimum length 20 octets), a UDP header (8 octets)
and whatever MAC level header is required by the media in use. The
theoretical maximum frame size is determined by the size of the
length field in the IP header. In almost all cases the actual
maximum and minimum sizes are determined by the limitations of the
media.
In theory it would be ideal to distribute the frame sizes in a way
that would evenly distribute the theoretical frame rates. These
recommendations incorporate this theory but specify frame sizes which
are easy to understand and remember. In addition, many of the same
frame sizes are specified on each of the media types to allow for
easy performance comparisons.
Note: The inclusion of an unrealistically small frame size on some of
the media types (i.e. with little or no space for data) is to help
characterize the per-frame processing overhead of the DUT.
9.1 Frame sizes to be used on Ethernet
64, 128, 256, 512, 1024, 1280, 1518
These sizes include the maximum and minimum frame sizes permitted
by the Ethernet standard and a selection of sizes between these
extremes with a finer granularity for the smaller frame sizes and
higher frame rates.
Bradner & McQuaid Informational [Page 5]RFC 1944 Benchmarking Methodology May 1996
9.2 Frame sizes to be used on 4Mb and 16Mb token ring
54, 64, 128, 256, 1024, 1518, 2048, 4472
The frame size recommendations for token ring assume that there is
no RIF field in the frames of routed protocols. A RIF field would
be present in any direct source route bridge performance test.
The minimum size frame for UDP on token ring is 54 octets. The
maximum size of 4472 octets is recommended for 16Mb token ring
instead of the theoretical size of 17.9Kb because of the size
limitations imposed by many token ring interfaces. The reminder
of the sizes are selected to permit direct comparisons with other
types of media. An IP (i.e. not UDP) frame may be used in
addition if a higher data rate is desired, in which case the
minimum frame size is 46 octets.
9.3 Frame sizes to be used on FDDI
54, 64, 128, 256, 1024, 1518, 2048, 4472
The minimum size frame for UDP on FDDI is 53 octets, the minimum
size of 54 is recommended to allow direct comparison to token ring
performance. The maximum size of 4472 is recommended instead of
the theoretical maximum size of 4500 octets to permit the same
type of comparison. An IP (i.e. not UDP) frame may be used in
addition if a higher data rate is desired, in which case the
minimum frame size is 45 octets.
9.4 Frame sizes in the presence of disparate MTUs
When the interconnect DUT supports connecting links with disparate
MTUs, the frame sizes for the link with the *larger* MTU SHOULD be
used, up to the limit of the protocol being tested. If the
interconnect DUT does not support the fragmenting of frames in the
presence of MTU mismatch, the forwarding rate for that frame size
shall be reported as zero.
For example, the test of IP forwarding with a bridge or router
that joins FDDI and Ethernet should use the frame sizes of FDDI
when going from the FDDI to the Ethernet link. If the bridge does
not support IP fragmentation, the forwarding rate for those frames
too large for Ethernet should be reported as zero.
10. Verifying received frames
The test equipment SHOULD discard any frames received during a test
run that are not actual forwarded test frames. For example, keep-
alive and routing update frames SHOULD NOT be included in the count
Bradner & McQuaid Informational [Page 6]RFC 1944 Benchmarking Methodology May 1996
of received frames. In any case, the test equipment SHOULD verify
the length of the received frames and check that they match the
expected length.
Preferably, the test equipment SHOULD include sequence numbers in the
transmitted frames and check for these numbers on the received
frames. If this is done, the reported results SHOULD include in
addition to the number of frames dropped, the number of frames that
were received out of order, the number of duplicate frames received
and the number of gaps in the received frame numbering sequence.
This functionality is required for some of the described tests.
11. Modifiers
It might be useful to know the DUT performance under a number of
conditions; some of these conditions are noted below. The reported
results SHOULD include as many of these conditions as the test
equipment is able to generate. The suite of tests SHOULD be first
run without any modifying conditions and then repeated under each of
the conditions separately. To preserve the ability to compare the
results of these tests any frames that are required to generate the
modifying conditions (management queries for example) will be
included in the same data stream as the normal test frames in place
of one of the test frames and not be supplied to the DUT on a
separate network port.
11.1 Broadcast frames
In most router designs special processing is required when frames
addressed to the hardware broadcast address are received. In
bridges (or in bridge mode on routers) these broadcast frames must
be flooded to a number of ports. The stream of test frames SHOULD
be augmented with 1% frames addressed to the hardware broadcast
address. The frames sent to the broadcast address should be of a
type that the router will not need to process. The aim of this
test is to determine if there is any effect on the forwarding rate
of the other data in the stream. The specific frames that should
be used are included in the test frame format document. The
broadcast frames SHOULD be evenly distributed throughout the data
stream, for example, every 100th frame.
The same test SHOULD be performed on bridge-like DUTs but in this
case the broadcast packets will be processed and flooded to all
outputs.
It is understood that a level of broadcast frames of 1% is much
higher than many networks experience but, as in drug toxicity
evaluations, the higher level is required to be able to gage the
Bradner & McQuaid Informational [Page 7]RFC 1944 Benchmarking Methodology May 1996
effect which would otherwise often fall within the normal
variability of the system performance. Due to design factors some
test equipment will not be able to generate a level of alternate
frames this low. In these cases the percentage SHOULD be as small
as the equipment can provide and that the actual level be
described in the report of the test results.
11.2 Management frames
Most data networks now make use of management protocols such as
SNMP. In many environments there can be a number of management
stations sending queries to the same DUT at the same time.
The stream of test frames SHOULD be augmented with one management
query as the first frame sent each second during the duration of
the trial. The result of the query must fit into one response
frame. The response frame SHOULD be verified by the test
equipment. One example of the specific query frame that should be
used is shown in Appendix C.
11.3 Routing update frames
The processing of dynamic routing protocol updates could have a
significant impact on the ability of a router to forward data
frames. The stream of test frames SHOULD be augmented with one
routing update frame transmitted as the first frame transmitted
during the trial. Routing update frames SHOULD be sent at the
rate specified in Appendix C for the specific routing protocol
being used in the test. Two routing update frames are defined in
Appendix C for the TCP/IP over Ethernet example. The routing
frames are designed to change the routing to a number of networks
that are not involved in the forwarding of the test data. The
first frame sets the routing table state to "A", the second one
changes the state to "B". The frames MUST be alternated during
the trial.
The test SHOULD verify that the routing update was processed by
the DUT.
11.4 Filters
Filters are added to routers and bridges to selectively inhibit
the forwarding of frames that would normally be forwarded. This
is usually done to implement security controls on the data that is
accepted between one area and another. Different products have
different capabilities to implement filters.
Bradner & McQuaid Informational [Page 8]RFC 1944 Benchmarking Methodology May 1996
The DUT SHOULD be first configured to add one filter condition and
the tests performed. This filter SHOULD permit the forwarding of
the test data stream. In routers this filter SHOULD be of the
form:
forward input_protocol_address to output_protocol_address
In bridges the filter SHOULD be of the form:
forward destination_hardware_address
The DUT SHOULD be then reconfigured to implement a total of 25
filters. The first 24 of these filters SHOULD be of the form:
block input_protocol_address to output_protocol_address
The 24 input and output protocol addresses SHOULD not be any that
are represented in the test data stream. The last filter SHOULD
permit the forwarding of the test data stream. By "first" and
"last" we mean to ensure that in the second case, 25 conditions
must be checked before the data frames will match the conditions
that permit the forwarding of the frame. Of course, if the DUT
reorders the filters or does not use a linear scan of the filter
rules the effect of the sequence in which the filters are input is
properly lost.
The exact filters configuration command lines used SHOULD be
included with the report of the results.
11.4.1 Filter Addresses
Two sets of filter addresses are required, one for the single
filter case and one for the 25 filter case.
The single filter case should permit traffic from IP address
198.18.1.2 to IP address 198.19.65.2 and deny all other
traffic.
The 25 filter case should follow the following sequence.
deny aa.ba.1.1 to aa.ba.100.1
deny aa.ba.2.2 to aa.ba.101.2
deny aa.ba.3.3 to aa.ba.103.3
...
deny aa.ba.12.12 to aa.ba.112.12
allow aa.bc.1.2 to aa.bc.65.1
deny aa.ba.13.13 to aa.ba.113.13
deny aa.ba.14.14 to aa.ba.114.14
Bradner & McQuaid Informational [Page 9]RFC 1944 Benchmarking Methodology May 1996
...
deny aa.ba.24.24 to aa.ba.124.24
deny all else
All previous filter conditions should be cleared from the
router before this sequence is entered. The sequence is
selected to test to see if the router sorts the filter
conditions or accepts them in the order that they were entered.
Both of these procedures will result in a greater impact on
performance than will some form of hash coding.
12. Protocol addresses
It is easier to implement these tests using a single logical stream
of data, with one source protocol address and one destination
protocol address, and for some conditions like the filters described
above, a practical requirement. Networks in the real world are not
limited to single streams of data. The test suite SHOULD be first run
with a single protocol (or hardware for bridge tests) source and
destination address pair. The tests SHOULD then be repeated with
using a random destination address. While testing routers the
addresses SHOULD be random and uniformly distributed over a range of
256 networks and random and uniformly distributed over the full MAC
range for bridges. The specific address ranges to use for IP are
shown in Appendix C.
13. Route Set Up
It is not reasonable that all of the routing information necessary to
forward the test stream, especially in the multiple address case,
will be manually set up. At the start of each trial a routing update
MUST be sent to the DUT. This routing update MUST include all of the
network addresses that will be required for the trial. All of the
addresses SHOULD resolve to the same "next-hop". Normally this will
be the address of the receiving side of the test equipment. This
routing update will have to be repeated at the interval required by
the routing protocol being used. An example of the format and
repetition interval of the update frames is given in Appendix C.
14. Bidirectional traffic
Normal network activity is not all in a single direction. To test
the bidirectional performance of a DUT, the test series SHOULD be run
with the same data rate being offered from each direction. The sum of
the data rates should not exceed the theoretical limit for the media.
Bradner & McQuaid Informational [Page 10]RFC 1944 Benchmarking Methodology May 1996
15. Single stream path
The full suite of tests SHOULD be run along with whatever modifier
conditions that are relevant using a single input and output network
port on the DUT. If the internal design of the DUT has multiple
distinct pathways, for example, multiple interface cards each with
multiple network ports, then all possible types of pathways SHOULD be
tested separately.
16. Multi-port
Many current router and bridge products provide many network ports in
the same module. In performing these tests first half of the ports
are designated as "input ports" and half are designated as "output
ports". These ports SHOULD be evenly distributed across the DUT
architecture. For example if a DUT has two interface cards each of
which has four ports, two ports on each interface card are designated
as input and two are designated as output. The specified tests are
run using the same data rate being offered to each of the input
ports. The addresses in the input data streams SHOULD be set so that
a frame will be directed to each of the output ports in sequence so
that all "output" ports will get an even distribution of packets from
this input. The same configuration MAY be used to perform a
bidirectional multi-stream test. In this case all of the ports are
considered both input and output ports and each data stream MUST
consist of frames addressed to all of the other ports.
Consider the following 6 port DUT:
--------------
---------| in A out X|--------
---------| in B out Y|--------
---------| in C out Z|--------
--------------
The addressing of the data streams for each of the inputs SHOULD be:
stream sent to input A:
packet to out X, packet to out Y, packet to out Z
stream sent to input B:
packet to out X, packet to out Y, packet to out Z
stream sent to input C
packet to out X, packet to out Y, packet to out Z
Note that these streams each follow the same sequence so that 3
packets will arrive at output X at the same time, then 3 packets at
Y, then 3 packets at Z. This procedure ensures that, as in the real
world, the DUT will have to deal with multiple packets addressed to
Bradner & McQuaid Informational [Page 11]RFC 1944 Benchmarking Methodology May 1996
the same output at the same time.
17. Multiple protocols
This document does not address the issue of testing the effects of a
mixed protocol environment other than to suggest that if such tests
are wanted then frames SHOULD be distributed between all of the test
protocols. The distribution MAY approximate the conditions on the
network in which the DUT would be used.
18. Multiple frame sizes
This document does not address the issue of testing the effects of a
mixed frame size environment other than to suggest that if such tests
are wanted then frames SHOULD be distributed between all of the
listed sizes for the protocol under test. The distribution MAY
approximate the conditions on the network in which the DUT would be
used. The authors do not have any idea how the results of such a test
would be interpreted other than to directly compare multiple DUTs in
some very specific simulated network.
19. Testing performance beyond a single DUT.
In the performance testing of a single DUT, the paradigm can be
described as applying some input to a DUT and monitoring the output.
The results of which can be used to form a basis of characterization
of that device under those test conditions.
This model is useful when the test input and output are homogenous
(e.g., 64-byte IP, 802.3 frames into the DUT; 64 byte IP, 802.3
frames out), or the method of test can distinguish between dissimilar
input/output. (E.g., 1518 byte IP, 802.3 frames in; 576 byte,
fragmented IP, X.25 frames out.)
By extending the single DUT test model, reasonable benchmarks
regarding multiple DUTs or heterogeneous environments may be
collected. In this extension, the single DUT is replaced by a system
of interconnected network DUTs. This test methodology would support
the benchmarking of a variety of device/media/service/protocol
combinations. For example, a configuration for a LAN-to-WAN-to-LAN
test might be:
(1) 802.3-> DUT 1 -> X.25 @ 64kbps -> DUT 2 -> 802.3
Or a mixed LAN configuration might be:
(2) 802.3 -> DUT 1 -> FDDI -> DUT 2 -> FDDI -> DUT 3 -> 802.3
Bradner & McQuaid Informational [Page 12]
RFC 1944 Benchmarking Methodology May 1996
In both examples 1 and 2, end-to-end benchmarks of each system could
be empirically ascertained. Other behavior may be characterized
through the use of intermediate devices. In example 2, the
configuration may be used to give an indication of the FDDI to FDDI
capability exhibited by DUT 2.
Because multiple DUTs are treated as a single system, there are
limitations to this methodology. For instance, this methodology may
yield an aggregate benchmark for a tested system. That benchmark
alone, however, may not necessarily reflect asymmetries in behavior
between the DUTs, latencies introduce by other apparatus (e.g.,
CSUs/DSUs, switches), etc.
Further, care must be used when comparing benchmarks of different
systems by ensuring that the DUTs' features/configuration of the
tested systems have the appropriate common denominators to allow
comparison.
20. Maximum frame rate
The maximum frame rates that should be used when testing LAN
connections SHOULD be the listed theoretical maximum rate for the
frame size on the media.
The maximum frame rate that should be used when testing WAN
connections SHOULD be greater than the listed theoretical maximum
rate for the frame size on that speed connection. The higher rate
for WAN tests is to compensate for the fact that some vendors employ
various forms of header compression.
A list of maximum frame rates for LAN connections is included in
Appendix B.
21. Bursty traffic
It is convenient to measure the DUT performance under steady state
load but this is an unrealistic way to gauge the functioning of a DUT
since actual network traffic normally consists of bursts of frames.
Some of the tests described below SHOULD be performed with both
steady state traffic and with traffic consisting of repeated bursts
of frames. The frames within a burst are transmitted with the
minimum legitimate inter-frame gap.
The objective of the test is to determine the minimum interval
between bursts which the DUT can process with no frame loss. During
each test the number of frames in each burst is held constant and the
inter-burst interval varied. Tests SHOULD be run with burst sizes of
16, 64, 256 and 1024 frames.
Bradner & McQuaid Informational [Page 13]
RFC 1944 Benchmarking Methodology May 1996
22. Frames per token
Although it is possible to configure some token ring and FDDI
interfaces to transmit more than one frame each time that the token
is received, most of the network devices currently available transmit
only one frame per token. These tests SHOULD first be performed
while transmitting only one frame per token.
Some current high-performance workstation servers do transmit more
than one frame per token on FDDI to maximize throughput. Since this
may be a common feature in future workstations and servers,
interconnect devices with FDDI interfaces SHOULD be tested with 1, 4,
8, and 16 frames per token. The reported frame rate SHOULD be the
average rate of frame transmission over the total trial period.
23. Trial description
A particular test consists of multiple trials. Each trial returns
one piece of information, for example the loss rate at a particular
input frame rate. Each trial consists of a number of phases:
a) If the DUT is a router, send the routing update to the "input"
port and pause two seconds to be sure that the routing has settled.
b) Send the "learning frames" to the "output" port and wait 2
seconds to be sure that the learning has settled. Bridge learning
frames are frames with source addresses that are the same as the
destination addresses used by the test frames. Learning frames for
other protocols are used to prime the address resolution tables in
the DUT. The formats of the learning frame that should be used are
shown in the Test Frame Formats document.
c) Run the test trial.
d) Wait for two seconds for any residual frames to be received.
e) Wait for at least five seconds for the DUT to restabilize.
24. Trial duration
The aim of these tests is to determine the rate continuously
supportable by the DUT. The actual duration of the test trials must
be a compromise between this aim and the duration of the benchmarking
test suite. The duration of the test portion of each trial SHOULD be
at least 60 seconds. The tests that involve some form of "binary
search", for example the throughput test, to determine the exact
result MAY use a shorter trial duration to minimize the length of the
search procedure, but the final determination SHOULD be made with
Bradner & McQuaid Informational [Page 14]RFC 1944 Benchmarking Methodology May 1996
full length trials.
25. Address resolution
The DUT SHOULD be able to respond to address resolution requests sent
by the DUT wherever the protocol requires such a process.
26. Benchmarking tests:
Note: The notation "type of data stream" refers to the above
modifications to a frame stream with a constant inter-frame gap, for
example, the addition of traffic filters to the configuration of the
DUT.
26.1 Throughput
Objective:
To determine the DUT throughput as defined in RFC 1242.
Procedure:
Send a specific number of frames at a specific rate through the
DUT and then count the frames that are transmitted by the DUT. If
the count of offered frames is equal to the count of received
frames, the rate of the offered stream is raised and the test
rerun. If fewer frames are received than were transmitted, the
rate of the offered stream is reduced and the test is rerun.
The throughput is the fastest rate at which the count of test
frames transmitted by the DUT is equal to the number of test
frames sent to it by the test equipment.
Reporting format:
The results of the throughput test SHOULD be reported in the form
of a graph. If it is, the x coordinate SHOULD be the frame size,
the y coordinate SHOULD be the frame rate. There SHOULD be at
least two lines on the graph. There SHOULD be one line showing
the theoretical frame rate for the media at the various frame
sizes. The second line SHOULD be the plot of the test results.
Additional lines MAY be used on the graph to report the results
for each type of data stream tested. Text accompanying the graph
SHOULD indicate the protocol, data stream format, and type of
media used in the tests.
We assume that if a single value is desired for advertising
purposes the vendor will select the rate for the minimum frame
size for the media. If this is done then the figure MUST be
expressed in frames per second. The rate MAY also be expressed in
bits (or bytes) per second if the vendor so desires. The
Bradner & McQuaid Informational [Page 15]RFC 1944 Benchmarking Methodology May 1996
statement of performance MUST include a/ the measured maximum
frame rate, b/ the size of the frame used, c/ the theoretical
limit of the media for that frame size, and d/ the type of
protocol used in the test. Even if a single value is used as part
of the advertising copy, the full table of results SHOULD be
included in the product data sheet.
26.2 Latency
Objective:
To determine the latency as defined in RFC 1242.
Procedure:
First determine the throughput for DUT at each of the listed frame
sizes. Send a stream of frames at a particular frame size through
the DUT at the determined throughput rate to a specific
destination. The stream SHOULD be at least 120 seconds in
duration. An identifying tag SHOULD be included in one frame
after 60 seconds with the type of tag being implementation
dependent. The time at which this frame is fully transmitted is
recorded (timestamp A). The receiver logic in the test equipment
MUST recognize the tag information in the frame stream and record
the time at which the tagged frame was received (timestamp B).
The latency is timestamp B minus timestamp A as per the relevant
definition frm RFC 1242, namely latency as defined for store and
forward devices or latency as defined for bit forwarding devices.
The test MUST be repeated at least 20 times with the reported
value being the average of the recorded values.
This test SHOULD be performed with the test frame addressed to the
same destination as the rest of the data stream and also with each
of the test frames addressed to a new destination network.
Reporting format:
The report MUST state which definition of latency (from RFC 1242)
was used for this test. The latency results SHOULD be reported
in the format of a table with a row for each of the tested frame
sizes. There SHOULD be columns for the frame size, the rate at
which the latency test was run for that frame size, for the media
types tested, and for the resultant latency values for each
type of data stream tested.
Bradner & McQuaid Informational [Page 16]RFC 1944 Benchmarking Methodology May 1996
26.3 Frame loss rate
Objective:
To determine the frame loss rate, as defined in RFC 1242, of a DUT
throughout the entire range of input data rates and frame sizes.
Procedure:
Send a specific number of frames at a specific rate through the
DUT to be tested and count the frames that are transmitted by the
DUT. The frame loss rate at each point is calculated using the
following equation:
( ( input_count - output_count ) * 100 ) / input_count
The first trial SHOULD be run for the frame rate that corresponds
to 100% of the maximum rate for the frame size on the input media.
Repeat the procedure for the rate that corresponds to 90% of the
maximum rate used and then for 80% of this rate. This sequence
SHOULD be continued (at reducing 10% intervals) until there are
two successive trials in which no frames are lost. The maximum
granularity of the trials MUST be 10% of the maximum rate, a finer
granularity is encouraged.
Reporting format:
The results of the frame loss rate test SHOULD be plotted as a
graph. If this is done then the X axis MUST be the input frame
rate as a percent of the theoretical rate for the media at the
specific frame size. The Y axis MUST be the percent loss at the
particular input rate. The left end of the X axis and the bottom
of the Y axis MUST be 0 percent; the right end of the X axis and
the top of the Y axis MUST be 100 percent. Multiple lines on the
graph MAY used to report the frame loss rate for different frame
sizes, protocols, and types of data streams.
Note: See section 18 for the maximum frame rates that SHOULD be
used.
26.4 Back-to-back frames
Objective:
To characterize the ability of a DUT to process back-to-back
frames as defined in RFC 1242.
Procedure:
Send a burst of frames with minimum inter-frame gaps to the DUT
and count the number of frames forwarded by the DUT. If the count
of transmitted frames is equal to the number of frames forwarded
the length of the burst is increased and the test is rerun. If
Bradner & McQuaid Informational [Page 17]RFC 1944 Benchmarking Methodology May 1996
the number of forwarded frames is less than the number
transmitted, the length of the burst is reduced and the test is
rerun.
The back-to-back value is the number of frames in the longest
burst that the DUT will handle without the loss of any frames.
The trial length MUST be at least 2 seconds and SHOULD be
repeated at least 50 times with the average of the recorded values
being reported.
Reporting format:
The back-to-back results SHOULD be reported in the format of a
table with a row for each of the tested frame sizes. There SHOULD
be columns for the frame size and for the resultant average frame
count for each type of data stream tested. The standard deviation
for each measurement MAY also be reported.
26.5 System recovery
Objective:
To characterize the speed at which a DUT recovers from an overload
condition.
Procedure:
First determine the throughput for a DUT at each of the listed
frame sizes.
Send a stream of frames at a rate 110% of the recorded throughput
rate or the maximum rate for the media, whichever is lower, for at
least 60 seconds. At Timestamp A reduce the frame rate to 50% of
the above rate and record the time of the last frame lost
(Timestamp B). The system recovery time is determined by
subtracting Timestamp B from Timestamp A. The test SHOULD be
repeated a number of times and the average of the recorded values
being reported.
Reporting format:
The system recovery results SHOULD be reported in the format of a
table with a row for each of the tested frame sizes. There SHOULD
be columns for the frame size, the frame rate used as the
throughput rate for each type of data stream tested, and for the
measured recovery time for each type of data stream tested.
26.6 Reset
Objective:
To characterize the speed at which a DUT recovers from a device or
software reset.
Bradner & McQuaid Informational [Page 18]RFC 1944 Benchmarking Methodology May 1996
Procedure:
First determine the throughput for the DUT for the minimum frame
size on the media used in the testing.
Send a continuous stream of frames at the determined throughput
rate for the minimum sized frames. Cause a reset in the DUT.
Monitor the output until frames begin to be forwarded and record
the time that the last frame (Timestamp A) of the initial stream
and the first frame of the new stream (Timestamp B) are received.
A power interruption reset test is performed as above except that
the power to the DUT should be interrupted for 10 seconds in place
of causing a reset.
This test SHOULD only be run using frames addressed to networks
directly connected to the DUT so that there is no requirement to
delay until a routing update is received.
The reset value is obtained by subtracting Timestamp A from
Timestamp B.
Hardware and software resets, as well as a power interruption
SHOULD be tested.
Reporting format:
The reset value SHOULD be reported in a simple set of statements,
one for each reset type.
27. Security Considerations
Security issues are not addressed in this document.
Bradner & McQuaid Informational [Page 19]RFC 1944 Benchmarking Methodology May 1996
28. Editors' Addresses
Scott Bradner
Harvard University
1350 Mass. Ave, room 813
Cambridge, MA 02138
Phone +1 617 495-3864
Fax +1 617 496-8500
EMail: sob@harvard.edu
Jim McQuaid
Bay Networks
3 Federal Street
Billerica, MA 01821
Phone +1 508 436-3915
Fax: +1 508 670-8145
EMail: jmcquaid@baynetworks.com
Bradner & McQuaid Informational [Page 20]
RFC 1944 Benchmarking Methodology May 1996
Appendix A: Testing Considerations
A.1 Scope Of This Appendix
This appendix discusses certain issues in the benchmarking
methodology where experience or judgment may play a role in the tests
selected to be run or in the approach to constructing the test with a
particular DUT. As such, this appendix MUST not be read as an
amendment to the methodology described in the body of this document
but as a guide to testing practice.
1. Typical testing practice has been to enable all protocols to be
tested and conduct all testing with no further configuration of
protocols, even though a given set of trials may exercise only one
protocol at a time. This minimizes the opportunities to "tune" a
DUT for a single protocol.
2. The least common denominator of the available filter functions
should be used to ensure that there is a basis for comparison
between vendors. Because of product differences, those conducting
and evaluating tests must make a judgment about this issue.
3. Architectural considerations may need to be considered. For
example, first perform the tests with the stream going between
ports on the same interface card and the repeat the tests with the
stream going into a port on one interface card and out of a port
on a second interface card. There will almost always be a best
case and worst case configuration for a given DUT architecture.
4. Testing done using traffic streams consisting of mixed protocols
has not shown much difference between testing with individual
protocols. That is, if protocol A testing and protocol B testing
give two different performance results, mixed protocol testing
appears to give a result which is the average of the two.
5. Wide Area Network (WAN) performance may be tested by setting up
two identical devices connected by the appropriate short- haul
versions of the WAN modems. Performance is then measured between
a LAN interface on one DUT to a LAN interface on the other DUT.
The maximum frame rate to be used for LAN-WAN-LAN configurations is a
judgment that can be based on known characteristics of the overall
system including compression effects, fragmentation, and gross link
speeds. Practice suggests that the rate should be at least 110% of
the slowest link speed. Substantive issues of testing compression
itself are beyond the scope of this document.
Bradner & McQuaid Informational [Page 21]RFC 1944 Benchmarking Methodology May 1996
Appendix B: Maximum frame rates reference
(Provided by Roger Beeman, Cisco Systems)
Size Ethernet 16Mb Token Ring FDDI
(bytes) (pps) (pps) (pps)
64 14880 24691 152439
128 8445 13793 85616
256 4528 7326 45620
512 2349 3780 23585
768 1586 2547 15903
1024 1197 1921 11996
1280 961 1542 9630
1518 812 1302 8138
Ethernet size
Preamble 64 bits
Frame 8 x N bits
Gap 96 bits
16Mb Token Ring size
SD 8 bits
AC 8 bits
FC 8 bits
DA 48 bits
SA 48 bits
RI 48 bits ( 06 30 00 12 00 30 )
SNAP
DSAP 8 bits
SSAP 8 bits
Control 8 bits
Vendor 24 bits
Type 16 bits
Data 8 x ( N - 18) bits
FCS 32 bits
ED 8 bits
FS 8 bits
Tokens or idles between packets are not included
FDDI size
Preamble 64 bits
SD 8 bits
FC 8 bits
DA 48 bits
SA 48 bits
SNAP
Bradner & McQuaid Informational [Page 22]RFC 1944 Benchmarking Methodology May 1996
DSAP 8 bits
SSAP 8 bits
Control 8 bits
Vendor 24 bits
Type 16 bits
Data 8 x ( N - 18) bits
FCS 32 bits
ED 4 bits
FS 12 bits
Appendix C: Test Frame Formats
This appendix defines the frame formats that may be used with these
tests. It also includes protocol specific parameters for TCP/IP over
Ethernet to be used with the tests as an example.
C.1. Introduction
The general logic used in the selection of the parameters and the
design of the frame formats is explained for each case within the
TCP/IP section. The same logic has been used in the other sections.
Comments are used in these sections only if there is a protocol
specific feature to be explained. Parameters and frame formats for
additional protocols can be defined by the reader by using the same
logic.
C.2. TCP/IP Information
The following section deals with the TCP/IP protocol suite.
C.2.1 Frame Type.
An application level datagram echo request is used for the test
data frame in the protocols that support such a function. A
datagram protocol is used to minimize the chance that a router
might expect a specific session initialization sequence, as might
be the case for a reliable stream protocol. A specific defined
protocol is used because some routers verify the protocol field
and refuse to forward unknown protocols.
For TCP/IP a UDP Echo Request is used.
C.2.2 Protocol Addresses
Two sets of addresses must be defined: first the addresses
assigned to the router ports, and second the address that are to
be used in the frames themselves and in the routing updates.
The network addresses 192.18.0.0 through 192.19.255.255 are have
been assigned to the BMWG by the IANA for this purpose. This
Bradner & McQuaid Informational [Page 23]RFC 1944 Benchmarking Methodology May 1996
assignment was made to minimize the chance of conflict in case a
testing device were to be accidentally connected to part of the
Internet. The specific use of the addresses is detailed below.
C.2.2.1 Router port protocol addresses
Half of the ports on a multi-port router are referred to as
"input" ports and the other half as "output" ports even though
some of the tests use all ports both as input and output. A
contiguous series of IP Class C network addresses from
198.18.1.0 to 198.18.64.0 have been assigned for use on the
"input" ports. A second series from 198.19.1.0 to 198.19.64.0
have been assigned for use on the "output" ports. In all cases
the router port is node 1 on the appropriate network. For
example, a two port DUT would have an IP address of 198.18.1.1
on one port and 198.19.1.1 on the other port.
Some of the tests described in the methodology memo make use of
an SNMP management connection to the DUT. The management
access address for the DUT is assumed to be the first of the
"input" ports (198.18.1.1).
C.2.2.2 Frame addresses
Some of the described tests assume adjacent network routing
(the reboot time test for example). The IP address used in the
test frame is that of node 2 on the appropriate Class C
network. (198.19.1.2 for example)
If the test involves non-adjacent network routing the phantom
routers are located at node 10 of each of the appropriate Class
C networks. A series of Class C network addresses from
198.18.65.0 to 198.18.254.0 has been assigned for use as the
networks accessible through the phantom routers on the "input"
side of DUT. The series of Class C networks from 198.19.65.0
to 198.19.254.0 have been assigned to be used as the networks
visible through the phantom routers on the "output" side of the
DUT.
C.2.3 Routing Update Frequency
The update interval for each routing protocol is may have to be
determined by the specifications of the individual protocol. For
IP RIP, Cisco IGRP and for OSPF a routing update frame or frames
should precede each stream of test frames by 5 seconds. This
frequency is sufficient for trial durations of up to 60 seconds.
Routing updates must be mixed with the stream of test frames if
longer trial periods are selected. The frequency of updates
should be taken from the following table.
Bradner & McQuaid Informational [Page 24]RFC 1944 Benchmarking Methodology May 1996
IP-RIP 30 sec
IGRP 90 sec
OSPF 90 sec
C.2.4 Frame Formats - detailed discussion
C.2.4.1 Learning Frame
In most protocols a procedure is used to determine the mapping
between the protocol node address and the MAC address. The
Address Resolution Protocol (ARP) is used to perform this
function in TCP/IP. No such procedure is required in XNS or
IPX because the MAC address is used as the protocol node
address.
In the ideal case the tester would be able to respond to ARP
requests from the DUT. In cases where this is not possible an
ARP request should be sent to the router's "output" port. This
request should be seen as coming from the immediate destination
of the test frame stream. (i.e. the phantom router (Figure 2)
or the end node if adjacent network routing is being used.) It
is assumed that the router will cache the MAC address of the
requesting device. The ARP request should be sent 5 seconds
before the test frame stream starts in each trial. Trial
lengths of longer than 50 seconds may require that the router
be configured for an extended ARP timeout.
+--------+ +------------+
| | | phantom |------ P LAN
A
IN A------| DUT |------------| |------ P LAN
B
| | OUT A | router |------ P LAN
C
+--------+ +------------+
Figure 2
In the case where full routing is being used
C.2.4.2 Routing Update Frame
If the test does not involve adjacent net routing the tester
must supply proper routing information using a routing update.
A single routing update is used before each trial on each
"destination" port (see section C.24). This update includes
the network addresses that are reachable through a phantom
router on the network attached to the port. For a full mesh
test, one destination network address is present in the routing
update for each of the "input" ports. The test stream on each
Bradner & McQuaid Informational [Page 25]RFC 1944 Benchmarking Methodology May 1996
"input" port consists of a repeating sequence of frames, one to
each of the "output" ports.
C.2.4.3 Management Query Frame
The management overhead test uses SNMP to query a set of
variables that should be present in all DUTs that support SNMP.
The variables for a single interface only are read by an NMS
at the appropriate intervals. The list of variables to
retrieve follow:
sysUpTime
ifInOctets
ifOutOctets
ifInUcastPkts
ifOutUcastPkts
C.2.4.4 Test Frames
The test frame is an UDP Echo Request with enough data to fill
out the required frame size. The data should not be all bits
off or all bits on since these patters can cause a "bit
stuffing" process to be used to maintain clock synchronization
on WAN links. This process will result in a longer frame than
was intended.
C.2.4.5 Frame Formats - TCP/IP on Ethernet
Each of the frames below are described for the 1st pair of DUT
ports, i.e. "input" port #1 and "output" port #1. Addresses
must be changed if the frame is to be used for other ports.
C.2.6.1 Learning Frame
ARP Request on Ethernet
-- DATAGRAM HEADER
offset data (hex) description
00 FF FF FF FF FF FF dest MAC address send to
broadcast address
06 xx xx xx xx xx xx set to source MAC address
12 08 06 ARP type
14 00 01 hardware type Ethernet = 1
16 08 00 protocol type IP = 800
18 06 hardware address length 48 bits
on Ethernet
19 04 protocol address length 4 octets
for IP
20 00 01 opcode request = 1
22 xx xx xx xx xx xx source MAC address
28 xx xx xx xx source IP address
Bradner & McQuaid Informational [Page 26]RFC 1944 Benchmarking Methodology May 1996
32 FF FF FF FF FF FF requesting DUT's MAC address
38 xx xx xx xx DUT's IP address
C.2.6.2 Routing Update Frame
-- DATAGRAM HEADER
offset data (hex) description
00 FF FF FF FF FF FF dest MAC address is broadcast
06 xx xx xx xx xx xx source hardware address
12 08 00 type
-- IP HEADER
14 45 IP version - 4, header length (4
byte units) - 5
15 00 service field
16 00 EE total length
18 00 00 ID
20 40 00 flags (3 bits) 4 (do not
fragment),
fragment offset-0
22 0A TTL
23 11 protocol - 17 (UDP)
24 C4 8D header checksum
26 xx xx xx xx source IP address
30 xx xx xx destination IP address
33 FF host part = FF for broadcast
-- UDP HEADER
34 02 08 source port 208 = RIP
36 02 08 destination port 208 = RIP
38 00 DA UDP message length
40 00 00 UDP checksum
-- RIP packet
42 02 command = response
43 01 version = 1
44 00 00 0
-- net 1
46 00 02 family = IP
48 00 00 0
50 xx xx xx net 1 IP address
53 00 net not node
54 00 00 00 00 0
58 00 00 00 00 0
62 00 00 00 07 metric 7
-- net 2
Bradner & McQuaid Informational [Page 27]RFC 1944 Benchmarking Methodology May 1996
66 00 02 family = IP
68 00 00 0
70 xx xx xx net 2 IP address
73 00 net not node
74 00 00 00 00 0
78 00 00 00 00 0
82 00 00 00 07 metric 7
-- net 3
86 00 02 family = IP
88 00 00 0
90 xx xx xx net 3 IP address
93 00 net not node
94 00 00 00 00 0
98 00 00 00 00 0
102 00 00 00 07 metric 7
-- net 4
106 00 02 family = IP
108 00 00 0
110 xx xx xx net 4 IP address
113 00 net not node
114 00 00 00 00 0
118 00 00 00 00 0
122 00 00 00 07 metric 7
-- net 5
126 00 02 family = IP
128 00 00 0
130 00 net 5 IP address
133 00 net not node
134 00 00 00 00 0
138 00 00 00 00 0
142 00 00 00 07 metric 7
-- net 6
146 00 02 family = IP
148 00 00 0
150 xx xx xx net 6 IP address
153 00 net not node
154 00 00 00 00 0
158 00 00 00 00 0
162 00 00 00 07 metric 7
C.2.4.6 Management Query Frame
To be defined.
Bradner & McQuaid Informational [Page 28]RFC 1944 Benchmarking Methodology May 1996
C.2.6.4 Test Frames
UDP echo request on Ethernet
-- DATAGRAM HEADER
offset data (hex) description
00 xx xx xx xx xx xx set to dest MAC address
06 xx xx xx xx xx xx set to source MAC address
12 08 00 type
-- IP HEADER
14 45 IP version - 4 header length 5 4
byte units
15 00 TOS
16 00 2E total length*
18 00 00 ID
20 00 00 flags (3 bits) - 0 fragment
offset-0
22 0A TTL
23 11 protocol - 17 (UDP)
24 C4 8D header checksum*
26 xx xx xx xx set to source IP address**
30 xx xx xx xx set to destination IP address**
-- UDP HEADER
34 C0 20 source port
36 00 07 destination port 07 = Echo
38 00 1A UDP message length*
40 00 00 UDP checksum
-- UDP DATA
42 00 01 02 03 04 05 06 07 some data***
50 08 09 0A 0B 0C 0D 0E 0F
* - change for different length frames
** - change for different logical streams
*** - fill remainder of frame with incrementing octets,
repeated if required by frame length
Bradner & McQuaid Informational [Page 29]RFC 1944 Benchmarking Methodology May 1996
Values to be used in Total Length and UDP message length fields:
frame size total length UDP message length
64 00 2E 00 1A
128 00 6E 00 5A
256 00 EE 00 9A
512 01 EE 01 9A
768 02 EE 02 9A
1024 03 EE 03 9A
1280 04 EE 04 9A
1518 05 DC 05 C8
Bradner & McQuaid Informational [Page 30]
