Network Working Group Hardev Soor
INTERNET-DRAFT Debra Stopp
Expiration Date: August 1999 Ixia Communications
Ralph Daniels
Netcom Systems
February 1999
Methodology for IP Multicast Benchmarking
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet- Drafts as reference
material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
Abstract
The purpose of this draft is to describe methodology specific to the
benchmarking of multicast IP forwarding devices. It builds upon the
tenets set forth in RFC 1944, RFC 2432 and other IETF Benchmarking
Methodology Working Group (BMWG) efforts. This document seeks to
extend these efforts to the multicast paradigm.
The BMWG produces two major classes of documents: Benchmarking
Terminology documents and Benchmarking Methodology documents. The
Terminology documents present the benchmarks and other related terms.
The Methodology documents define the procedures required to collect
the benchmarks cited in the corresponding Terminology documents.
Soor, Stopp, & Daniels [Page 1]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
1. Introduction
This document defines a specific set of tests that vendors can use to
measure and report the performance characteristics and forwarding
capabilities of network devices that support IP multicast protocols.
The results of these tests will provide the user comparable data from
different vendors with which to evaluate these devices.
A previous document, " Terminology for IP Multicast Benchmarking"
(RFC 2432), 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.
This methodology will focus on one source to many destinations,
although many of the tests described may be extended to use multiple
source to multiple destination IP multicast communication.
2. Key Words to Reflect Requirements
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119.
3. Test set up
Figure 1 shows a typical setup for an IP multicast test, with one
source to multiple destinations, although this MAY be extended to
multiple source to multiple destinations.
+----------------+
+------------+ | |
+--------+ | |--------->| destination(1) |
| | | | | |
| source |-------->| | +----------------+
| | | | +----------------+
+--------+ | D U T |--------->| |
| | | destination(2) |
| | | |
| | +----------------+
| | . . .
| | +----------------+
| | | |
| |--------->| destination(n) |
| | | |
| | +----------------+
| |
+------------+
Figure 1
Soor, Stopp, & Daniels [Page 2]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
Generally , the destination ports first join the desired number of
multicast groups by sending IGMP Join Group messages to the DUT.
To verify that all destination ports successfully joined the
appropriate groups, the source port MUST transmit IP multicast
frames destined for these groups.
In addition, all transmitted frames MUST contain a recognizable
pattern that can be filtered on in order to ensure the receipt of only
the frames that are involved in the test.
3.1 Test Considerations
3.1.1 IGMP Support
Each of the receiving ports should support and be able to test both IGMP
version 1 and IGMP version 2.
Each receiving port should be able to respond to IGMP queries during the test.
Each receiving port should also send LEAVE (running IGMP version 2) after each test.
3.1.2 Group Addresses
The Class D Group address should be changed between tests. Many
DUTs have memory or cache that is not cleared properly and can
bias the results.
3.1.3 Frame Sizes
Each test should be run with different Multicast Frame Sizes. The result to vary greatly.
The source frames should have a TTL value large enough to accommodate the DUT/SUT.
4. Forwarding and Throughput
This section contains the description of the tests that are related to the
characterization of the packet forwarding of a DUT/SUT in a multicast
environment. Some metrics extend the concept of throughput presented
in RFC 1242. The notion of Forwarding Rate is cited in RFC 2285
4.1 Mixed Class Throughput
Definition
The maximum rate at which none of the offered frames, comprised from a
unicast Class and a multicast Class, to be forwarded are dropped by the
Soor, Stopp, & Daniels [Page 3]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
device across a fixed number of ports.
Procedure
Multicast and unicast traffic are mixed together in the same aggregated
traffic stream in order to simulate the non-homogenous networking
environment. While the multicast traffic is transmitted from one source to
multiple destinations, the unicast traffic MAY be evenly distributed across
the DUT architecture. In addition, the DUT SHOULD learn the appropriate
unicast IP addresses, either by sending ARP frames from each unicast address,
sending a RIP packet or by assigning static entries into the DUT address
table.
The rates at which traffic is transmitted for both traffic classes MUST be
set up in one of two ways:
a) A percentage of the bandwidth is allocated for each traffic class and
frames for each class are transmitted at the rate equal to the allocated
bandwidth. For example, 64 byte frames can be transmitted at a theoretical
maximum rate of 148810 frames/second. If 80 percent of the bandwidth is
allocated for unicast traffic and 20 percent for multicast traffic, then
unicast traffic will be sent at a maximum rate of 119048 frames/second and
the multicast traffic at a rate of 29762 frames/second.
b) Transmission rate is fixed for both traffic classes and a percentage of
number of frames for each traffic class is specified. For example, if a
fixed rate of 100% of theoretical maximum is desired, then 64 byte frames
will be sent at 148810 frames/second for both unicast and multicast
traffic. If 80 percent of the frames are to be unicast and 20 percent
multicast, then for a duration of 10 seconds, 1190480 frames of unicast
and 297620 frames of multicast will be sent. This fixed rate scenario
actually over-subscribes the bandwidth, potentially causing congestion in
the DUT.
The transmission of the frames MUST be set up so that they form a
deterministic distribution while still maintaining the specified bandwidth
and transmission rates. See Appendix A for a discussion on determining an
even distribution.
Similar to the Frame loss rate test in RFC 1944, 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 HOULD 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.
Soor, Stopp, & Daniels [Page 4]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
Result
Transmit and Receive rates in frames per second for each source and
destination port for both unicast and multicast traffic for each trial
percent transmit rate. The result report SHOULD contain the number of
frames transmitted and received per port per class type (unicast and
multicast traffic), reported in number of frames and percent loss per port.
4.2 Scaled Group Forwarding Matrix
Definition:
A table that demonstrates Forwarding Rate as a function of tested
multicast groups for a fixed number of tested DUT/SUT ports.
Procedure:
Multicast traffic is sent at a fixed percent of line rate with a fixed
number of receive ports at a fixed frame length.
The receive ports will join an initial number of groups and the sender
will transmit to the same groups after a certain delay (a few seconds).
Then the receive ports will join an incremental value of groups and the
transmit port will send to all groups joined (initial plus incremental).
The receive ports will continue joining in the incremental fashion until a
user defined maximum is reached.
Results:
For each group load the result WILL display frame rate, frames transmitted,
total frames received, total frames loss, and percent loss. The frame loss
per receive port per group SHOULD also be available.
4.3 Aggregated Multicast Throughput
Definition:
The maximum rate at which none of the offered frames to be
forwarded through N destination interfaces of the same multicast
group are dropped.
Procedure:
Multicast traffic is sent at a fixed percent of line rate with a fixed
number of groups at a fixed frame length for a fixed duration of time.
Soor, Stopp, & Daniels [Page 5]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
The initial number of receive ports will join the group(s) and the
sender will transmit to the same groups after a certain delay (a few
seconds).
Then the an incremental number of receive ports will join the same groups
and then the Multicast traffic is sent as stated.
The receive ports will continue to be added and the Multicast traffic sent
until a user defined maximum number of ports is reached.
Results:
For each number of receive ports the result WILL display frame rate, frames
transmitted, total frames received, total frames loss, and percent loss.
The frame loss per receive port per group SHOULD also be available.
4.4 Encapsulation (Tunneling) Throughput
This sub-section provides the description of tests that help in obtaining
throughput measurements when a DUT or a set of DUTs are acting as tunnel
endpoints. The following Figure 2 presents the scenario for the tests.
Client A DUT A Network DUT B Client B
---------- ----------
| | ------ | |
-------(a) (b)| |(c) ( ) (d)| |(e) (f)-------
||||||| -----> | |---->( )----->| |-----> |||||||
------- | | ------ | | -------
| | | |
---------- ----------
Figure 2
--------
A tunnel is created between DUT A (the encapsulator) and DUT B (the
decapsulator). Client A is acting as a source and Client B is the
destination. Client B joins a multicast group (for example, 224.0.1.1) and it
sends an IGMP Join message to DUT B to join that group. Client A now wants to
transmit some traffic to Client B. It will send the multicast traffic to DUT
A which encapsulates the multicast frames, sends it to DUT B which will
decapsulate the same frames and forward them to Client B.
4.4.1 Encapsulation Throughput
Soor, Stopp, & Daniels [Page 6]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
Definition
The maximum rate at which frames offered a DUT are encapsulated and correctly
forwarded by the DUT without loss.
Procedure
To test the forwarding rate of the DUT when it has to go through the process
of encapsulation, a test port B is injected at the other end of DUT A
(Figure B) that will receive the encapsulated frames and measure the
throughput. Also, a test port A is used to generate multicast frames that
will be passed through the tunnel.
The following is the test setup:
Test port A DUT A Test port B
---------- (c') (d')---------
| |-------------->| |
-------(a) (b)| | | |
||||||| -----> | | ------ ---------
------- | |(c) ( N/W )
| |---->( )
---------- ------
Figure 3
--------
In Figure 2, a tunnel is created with the local IP address of DUT A as the
beginning of the tunnel (point c) and the IP address of DUT B as the end
of the tunnel (point d). DUT B is assumed to have the tunneling protocol
enabled so that the frames can be decapsulated. When the test port B is
inserted in between the DUT A and DUT B (Figure 3), the endpoint of tunnel
has to be re-configured to be directed to the test port B's IP address. For
example, in Figure 3, point c' would be assigned as the beginning of the
tunnel and point d' as the end of the tunnel. The test port B is acting as
the end of the tunnel, and it does not have to support any tunneling protocol
since the frames do not have to be decapsulated. Instead, the received
encapsulated frames are used to calculate the throughput and other necessary
measurements.
Result
Throughput in frames per second for each destination port. The results
should also contain the number of frames transmitted and received per port.
4.4.2 Decapsulation Throughput
Soor, Stopp, & Daniels [Page 7]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
Definition
The maximum rate at which frames offered a DUT are decapsulated and correctly
forwarded by the DUT without loss.
Procedure
The decapsulation process returns the tunneled unicast frames back to their
multicast format. This test measures the throughput of the DUT when it has to
perform the process of decapsulation, therefore, a test port C is used at the
end of the tunnel to receive the decapsulated frames (Figure 4).
Test port A DUT A Test port B DUT B Test port C
---------- ----------
| | | |
-------(a) (b)| |(c) ------ (d)| |(e) (f)-------
||||||| -----> | |----> |||||| ----->| |-----> |||||||
------- | | ------ | | -------
| | | |
---------- ----------
Figure 4
--------
In Figure 4, the encapsulation process takes place in DUT A. This may
effect the throughput of the DUT B. Therefore, two test ports should be
used to separate the encapsulation and decapsulation processes. Client A
is replaced with the test port A which will generate a multicast frame that
will be encapsulated by DUT A. Another test port B is inserted between DUT
A and DUT B that will receive the encapsulated frames and forward it to DUT
B. Test port C will receive the decapsulated frames and measure the
throughput.
Result
Throughput in frames per second for each destination port . The results
should also contain the number of frames transmitted and received per port.
4.4.3 Re-encapsulation Throughput
Definition
The maximum rate at which frames of one encapsulated format offered a DUT are
converted to another encapsulated format and correctly forwarded by the DUT
without loss.
Procedure
Soor, Stopp, & Daniels [Page 8]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
Re-encapsulation takes place in DUT B after test port C has received the
decapsulated frames. These decapsulated frames will be re-inserted with
a new encapsulation frame and sent to test port B which will measure the
throughput. See Figure 5.
Test port A DUT A Test port B DUT B Test port C
---------- ----------
| | | |
-------(a) (b)| |(c) ------ (d)| |(e) (f)-------
||||||| -----> | |----> |||||| <---->| |<----> |||||||
------- | | ------ | | -------
| | | |
---------- ----------
Figure 5
--------
Result
Throughput in frames per second for each destination port. The results
should also contain the number of frames transmitted and received per port.
5. Forwarding Latency
This section presents methodologies relating to the characterization of
the forwarding latency of a DUT/SUT in a multicast environment. It
extends the concept of latency characterization presented in RFC 1944.
5.1 Multicast Latency
Definition
The set of individual latencies from a single input port on the DUT or SUT to
all tested ports belonging to the destination multicast group.
Procedure
According to RFC 1944, a tagged frame is sent half way through the
transmission that contains a timestamp used for calculation of latency. In
the multicast situation, a tagged frame is sent to all destinations for each
multicast group and latency calculated on a per multicast group basis. Note
that this test MUST be run using the transmission rate that is less than the
multicast throughput of the DUT.
Result
The latency value for each multicast group address.
Soor, Stopp, & Daniels [Page 9]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
5.2 Min/Max/Average Multicast Latency
Definition:
The difference between the maximum latency measurement and the
minimum latency measurement from the set of latencies produced by
the Multicast Latency benchmark.
Procedure:
For the entire duration of the Latency test the smallest latency, the
largest latency, the sum of latencies, and the number should be tracked
per receive port.
The test can also increment bucket counters that represent a range latency
range. This can be used to create a histogram. From the histogram, minimum,
maximum, and average the test results can show the jitter.
Results:
For each port the results WILL display the number of frames, minimum latency,
maximum latency, and the average latency. The results SHOULD also display
the histogram of latencies.
6. Overhead
This section presents methodology relating to the characterization of
the overhead delays associated with explicit operations found in
multicast environments.
6.1 Group Join Delay
Definition:
The time duration it takes a DUT to start forwarding multicast
packets from the time a successful IGMP group membership report
has been issued to the DUT.
Procedure:
Traffic is sent on the source port at the same time as the IGMP JOIN Group
message is transmitted from the destination ports. The join delay is the
difference in time from when the IGMP Join is sent and the first frame is
received.
One of the keys is to transmit at the fastest rate the DUT can handle multicast
frames. This is to get the best resultion in the Join Delay. However, you
Soor, Stopp, & Daniels [Page 10]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
do not want to transmit the frames to fast that frames are dropped by the DUT.
The test should be ran using both IGMPv1 JOIN and IGMPv2 JOIN.
Results:
The JOIN delay for each port.
6.2 Group Leave Delay
Definition
The time duration it takes a DUT to cease forwarding multicast packets after
a corresponding IGMP "Leave Group" message has been successfully offered to
the DUT.
Procedure
Traffic is sent on the source port at the same time as the IGMP Leave Group
messages are transmitted from the destination ports. The frames on both the
source and destination ports are sent with the timestamps inserted. The Group
Leave Delay is the difference in the value of the timestamp A of the first
IGMP Leave Group frame sent and the timestamp B of the last frame that is
received on that destination port.
Group Leave delay = timestamp B - timestamp A
Result
Group Leave Delay values for each multicast group address on each destination
port. Also, the number of frames transmitted and received, and percent loss
may be displayed.
8. Capacity
This section offers terms relating to the identification of multicast
group limits of a DUT/SUT.
8.1 Multicast Group Capacity
Definition:
The maximum number of multicast groups a SUT/DUT can support while
maintaining the ability to forward multicast frames to all
multicast groups registered to that SUT/DUT.
Procedure:
One or more receiving ports will join an initial number of groups. Then
Soor, Stopp, & Daniels [Page 11]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
after a delay the source port will transmit to each group at a transmission
rate that the DUT can handle. If all frames sent are forwarded and received
the receiving ports will join an incremental value of groups. Then after a
a delay the source port will transmit to all groups at a transmission rate
that the DUT can handle. If all frames sent are forwarded and received the
receiving ports will continuing joining and testing until a frame is not forwarded
nor received.
The group capacity resolution will be the incremental value. So the capacity
could be greater then last capacity passed but less then the one that failed.
Once a capacity is determined the test should be re run with greater delays after
the JOIN and a slower transmission rate. And the initial group level should be
raised to about five less then the previous capacity and incremental value should
be set to one.
Results:
The number of groups passed vs the number of groups failed. The results SHOULD
give details when the frame fails to be forwarded about how many frames did and
did not get forwarded. Which groups DID and DID NOT get forwarded.
9. Burdened Response
9.1 Burdened Response
Definition:
A measured response collected from a DUT/SUT in light of
interacting, or potentially interacting, distinct stimuli.
Procedure:
TBD
Results:
TBD
9.2 Forwarding Burdened Multicast Latency
Definition
A multicast latency taken from a DUT/SUT in the presence of a traffic
forwarding requirement.
Procedure
Soor, Stopp, & Daniels [Page 12]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
This test build upon the Multicast Latency as described in 1.5.1. This test
measures latencies for a set of offered load values and reports the forwarding
rates.
Result
A table of offered load (percent of maximum rate), forwarding rates and
latencies for each multicast group involved.
9.3 Forwarding Burdened Join Delay
Definition:
A multicast Group Join Delay taken from a DUT in the presence of a
traffic forwarding requirement.
Procedure:
TBD
Results:
TBD
Appendix A: Determining an even distribution
A.1 Scope Of This Appendix
This appendix discusses the suggested approach to configuring the deterministic distribution methodology for tests that involve both multicast and unicast traffic classes in an aggregated traffic stream. 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.
It is important to understand and fully define the distribution of frames among all multicast and unicast destinations. If the distribution is not well defined or understood, the throughput and forwarding metrics are not meaningful.
In a homogeneous environment, a large, single burst of multicast frames may be followed by a large burst of unicast frames. This is a very different distribution than that of a non-homogeneous environment, where the multicast and unicast frames are intermingled throughout the entire transmission.
The recommended distribution is that of the non-homogeneous environment because it more closely represents a real-world scenario. The distribution is modeled by calculating the number of multicast frames per destination port as a burst, then calculating the number of unicast frames to transmit as a percentage of the total frames transmitted. The overall effect of the distribution is small bursts of multicast frames intermingled with small bursts of unicast frames.
Example
Frame size = 64
Duration of test = 10 seconds
Transmission rate = 100% of maximum rate
Mapping for unicast traffic: Port 1 to Port 2
Port 3 to port 4
Mapping for multicast traffic: Port 1 to Ports 2,3,4
Number of Multicast group addresses per destination port = 3
Multicast groups joined by Port 2: 224.0.1.1
224.0.1.2
Soor, Stopp, & Daniels [Page 13]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
224,0.1.3
Multicast groups joined by Port 3: 224.0.1.4
224.0.1.5
224,0.1.6
Multicast groups joined by Port 4: 224.0.1.7
224.0.1.8
224,0.1.9
Percentage of Unicast frames = 20
Percentage of Multicast frames = 80
Total number of frames to be transmitted = 148810 fps * 10 sec
= 1488100 frames
Number of unicast frames = 20/100 * 1488100 = 297620 frames
Number of multicast frames = 80/100 * 1488100 = 1190480 frames
Unicast burst size = 20 * 9 = 180
Multicast burst size = 80 * 9 = 720
Loop counter = 1488100 / 900 = 1653.4444 (round it off to 1653)
Therefore, the actual number of frames that will be transmitted:
Unicast frames = 1653 * 180 = 297540 frames
Multicast frames = 1653 * 720 = 1190160 frames
The following pattern will be established:
UUUMMMMMMMMMMMMUUUMMMMMMMMMMMMUUUMMMMMMMMMMMMUUUMMMMMMMMMMMM
where U represents 60 Unicast frames (UUU = 180 frames)
M represents 60 Multicast frames (MMMMMMMMMMMM = 720 frames)
3. Security Considerations.
As this document is solely for the purpose of providing metric methodology
and describes neither a protocol nor a protocol's implementation, there
are no security considerations associated with this document.
5. References
[Br91] Bradner, S., "Benchmarking Terminology for Network
Interconnection Devices", RFC 1242, July 1991.
[Br96] Bradner, S., and J. McQuaid, "Benchmarking Methodology for
Network Interconnect Devices", RFC 1944, May 1996.
[Br97] Bradner, S. "Use of Keywords in RFCs to Reflect Requirement
Levels, RFC 2119, March 1997
Soor, Stopp, & Daniels [Page 14]
INTERNET-DRAFT IP Multicast Benchmarking Methodology February 1999
[Du98] Dubray, K., "Terminology for IP Multicast Benchmarking",
RFC 2432, October 1998.
[Hu95] Huitema, C. "Routing in the Internet." Prentice-Hall, 1995.
[Ka98] Kosiur, D., "IP Multicasting: the Complete Guide to Interactive
Corporate Networks", John Wiley & Sons, Inc, 1998.
[Ma98] Mandeville, R., "Benchmarking Terminology for LAN Switching
Devices", RFC 2285, February 1998.
[Mt98] Maufer, T. "Deploying IP Multicast in the Enterprise."
Prentice-Hall, 1998.
[Se98] Semeria, C. and Maufer, T. "Introduction to IP Multicast
Routing." http://www.3com.com/nsc/501303.html 3Com Corp.,
1998.
6. Author's Address
Hardev Soor
Ixia Communications
4505 Las Virgenes Road, Suite 209
Calabasas, CA 91302
USA
Phone: 818 871 1800
EMail: hardev@ixiacom.com
Debra Stopp
Ixia Communications
4505 Las Virgenes Road, Suite 209
Calabasas, CA 91302
USA
Phone: 818 871 1800
EMail: debby@ixiacom.com
Ralph Daniels
Netcom Systems
948 Loop Road
Clayton, NC 27520
USA
Phone: 919 550 9475
EMail: Ralph_Daniels@NetcomSystems.com
Soor, Stopp, & Daniels [Page 15]
