OSPF Convergence Bench: Benchmarking Basic OSPF Single Router Control Plane Convergence
A repeatable way to measure how quickly one router's OSPF process reacts to a change.
Manral, V., White, R., and A. Shaikh, "Benchmarking Basic OSPF Single Router Control Plane Convergence", RFC 4061, DOI 10.17487/RFC4061, April 2005, <https://www.rfc-editor.org/info/rfc4061>.
@misc{rfc4061,
series = {Request for Comments},
number = 4061,
howpublished = {RFC 4061},
publisher = {RFC Editor},
doi = {10.17487/RFC4061},
url = {https://www.rfc-editor.org/info/rfc4061},
author = {V. Manral and R. White and A. Shaikh},
title = {{Benchmarking Basic OSPF Single Router Control Plane Convergence}},
pagetotal = 16,
year = 2005,
month = apr,
} RFC text
Contents
Network Working Group V. Manral
Request for Comments: 4061 SiNett Corp.
Category: Informational R. White
Cisco Systems
A. Shaikh
AT&T Labs (Research)
April 2005
Benchmarking Basic OSPF Single Router Control Plane Convergence
Status of This Memo
This memo provides information for the Internet community. It does
not specify an Internet standard of any kind. Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2005).
Abstract
This document provides suggestions for measuring OSPF single router
control plane convergence. Its initial emphasis is on the control
plane of a single OSPF router. We do not address forwarding plane
performance.
NOTE: In this document, the word "convergence" relates to single
router control plane convergence only.
Table of Contents
1. Introduction....................................................2
2. Specification of Requirements...................................2
3. Overview and Scope..............................................3
4. Reference Topologies............................................4
5. Basic Performance Tests.........................................5
5.1. Time Required to Process an LSA...........................5
5.2. Flooding Time.............................................6
5.3. Shortest Path First Computation Time......................6
6. Basic Intra-area OSPF Tests.....................................8
6.1. Forming Adjacencies on Point-to-Point Links
(Initialization)..........................................9
6.2. Forming Adjacencies on Point-to-Point Links...............9
6.3. Forming Adjacencies with Information Already in the
Database.................................................10
6.4. Designated Router Election Time on a Broadcast Network...11
6.5. Initial Convergence Time on a Broadcast Network,
Test 1...................................................11
6.6. Initial Convergence Time on a Broadcast Network,
Test 2...................................................12
6.7. Link Down with Layer Two Detection.......................12
6.8. Link Down with Layer Three Detection.....................13
7. Security Considerations........................................13
8. Acknowledgements...............................................13
9. Normative References...........................................14
10. Informative References.........................................14
Authors' Addresses.................................................15
Full Copyright Statement...........................................16
1. Introduction
There is a growing interest in routing protocol convergence testing,
with many people looking at various tests to determine how long it
takes for a network to converge after various conditions occur. The
major problem with this sort of testing is that the framework of the
tests has a major impact on the results; for instance, determining
when a network is converged, what parts of the router's operation are
considered within the testing, and other such things will have a
major impact on the apparent performance that routing protocols
provide.
This document attempts to provide a framework for Open Shortest Path
First [OSPF] performance testing, and to provide some tests for
measuring some aspects of OSPF performance. The motivation of the
document is to provide a set of tests that can provide the user
comparable data from various vendors with which to evaluate the OSPF
protocol performance on the devices.
2. Specification of 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 [RFC2119]. RFC 2119
key words in this document are used to ensure methodological control,
which is very important in the specification of benchmarks. This
document does not specify a network-related protocol.
3. Overview and Scope
Although this document describes a specific set of tests aimed at
characterizing the single router control plane convergence
performance of OSPF processes in routers or other boxes that
incorporate OSPF functionality, a key objective is to propose
methodologies that produce directly comparable convergence-related
measurements.
The following considerations are outside the scope of this document:
o The interactions of convergence and forwarding; testing is
restricted to events occurring within the control plane.
Forwarding performance is the primary focus in [INTERCONNECT], and
it is expected to be dealt with in work that ensues from [FIB-
TERM].
o Inter-area route generation, AS-external route generation, and
simultaneous traffic on the control and data paths within the DUT.
Although the tests outlined in this document measure SPF time,
flooding times, and other aspects of OSPF convergence performance,
this document does not provide tests for measuring external or
summary route generation, route translation, or other OSPF inter-
area and external routing performance. These areas are expected
to be dealt with in a later document.
The tests should be run more than once, since a single test run
cannot be relied on to produce statistically sound results. The
number of test runs and any variations between the tests should be
recorded in the test results (see [TERM] for more information on
what items should be recorded in the test results).
4. Reference Topologies
Several reference topologies that are used throughout the tests are
described in the remaining sections of this document. All of the
topologies have been collectively placed in one section to avoid
repetition.
o Reference Topology 1 (Emulated Topology)
( )
DUT----Generator----( emulated topology )
( )
A simple back-to-back configuration. It's assumed that the link
between the generator and the DUT is a point-to-point link, while
the connections within the generator represent some emulated
topology.
o Reference Topology 2 (Generator and Collector)
( )
Collector-----DUT-----Generator--( emulated topology )
\ / ( )
\------------/
All routers are connected through point-to-point links. The cost
of all links is assumed to be the same unless otherwise noted.
o Reference Topology 3 (Broadcast Network)
DUT R1 R2
| | |
-+------+------+-----.....
Any number of routers could be included on the common broadcast
network.
o Reference Topology 4 (Parallel Links)
/--(link 1)-----\ ( )
DUT Generator--( emulated topology )
\--(link 2)-----/ ( )
In all cases the tests and topologies are designed to allow
performance measurements to be taken all on a single device, whether
this is the DUT or some other device in the network. This eliminates
the need for synchronized clocks within the test networks.
5. Basic Performance Tests
These tests will measure aspects of the OSPF implementation as a
process on the device under test, including
o time required to process an LSA,
o flooding time, and
o Shortest Path First computation.
5.1. Time Required to Process an LSA
o Using reference topology 1 (Emulated Topology), begin with all
links up and a full adjacency established between the DUT and the
generator.
Note: The generator does not have direct knowledge of the state of
the adjacency on the DUT. The fact that the adjacency may be in
Full state on the generator does not mean that the DUT is ready.
It may still (and is likely to) be requesting LSAs from the
generator. This process, involving processing of requested LSAs,
will affect the results of the test. The generator should either
wait until it sees the DUT's router-LSA listing the adjacency with
the generator or introduce a configurable delay before starting
the test.
o Send an LSA that is already in the DUT (a duplicate LSA), note the
time difference between when the LSA is sent and when the ack is
received. This measures the time taken to propagate the LSA and
the ack, as well as the processing time of the duplicate LSA.
This is dupLSAprocTime.
o Send a new LSA from the generator to the DUT, followed immediately
by a duplicate LSA (LSA that already resides in the database of
DUT, but not the same as the one just sent).
o The DUT will acknowledge this second LSA immediately; note the
time of this acknowledgement. This is newLSAprocTime.
The amount of time required for an OSPF implementation to process
the new LSA can be computed by subtracting dupLSAprocTime from
newLSAprocTime.
Note: The duplicate LSA cannot be the same as the one just sent
because of the MinLSInterval restriction [OSPF]. This test is
taken from [BLACKBOX].
Note: This time may or may not include the time required to
perform flooding-related operations, depending on when the
implementation sends the ack: before it floods the LSA further, or
after it does, or anywhere in between. In other words, this
measurement may not mean the same thing in all implementations.
5.2. Flooding Time
o Using reference topology 2 (Generator and Collector), enable OSPF
on all links and allow the devices to build full adjacencies.
Configure the collector so that it will block all flooding toward
the DUT (but so that it continues receiving advertisements from
the DUT).
o Inject a new set of LSAs from the generator toward the collector
and the DUT.
o On the collector, note the time the flooding is complete across
the link to the generator. Also note the time the flooding is
complete across the link from the DUT.
The time from when the last LSA is received on the collector from the
generator to when the last LSA is received on the collector from the
DUT should be measured during this test. This time is important in
link state protocols, since the loop-free nature of the network is
reliant on the speed at which revised topology information is
flooded.
Depending on the number of LSAs flooded, the sizes of the LSAs, the
number of LSUs, and the rate of flooding, these numbers could vary by
some amount. The settings and variances of these numbers should be
reported with the test results.
5.3. Shortest Path First Computation Time
o Use reference topology 1 (Emulated Topology), beginning with the
DUT and the generator fully adjacent.
o The default SPF timer on the DUT should be set to 0 so that any
new LSA that arrives immediately results in the SPF calculation
[BLACKBOX].
o The generator should inject a set of LSAs toward the DUT; the DUT
should be allowed to converge and install all best paths in the
local routing table, etc.
o Send an LSA that is already in the DUT (a duplicate LSA), note the
time difference between when the LSA is sent and when the ack is
received. This measures the time taken to propagate the LSA and
the ack, as well as the processing time of the duplicate LSA.
This is dupLSAprocTime.
o Change the link cost between the generator and the emulated
network it is advertising, and transmit the new LSA to the DUT.
o Immediately inject another LSA that is a duplicate of some other
LSA the generator has previously injected (preferably a stub
network someplace within the emulated network).
Note: The generator should make sure that outbound LSA packing is
not performed for the duplicate LSAs and that they are always sent
in a separate Link-state Update packet. Otherwise, if the LSA
carrying the topology change and the duplicate LSA are in the same
packet, the SPF starts after the duplicate LSA is acked.
o Measure the time between transmitting the second (duplicate) LSA
and the acknowledgement for that LSA; this is the totalSPFtime.
The total time required to run SPF can be computed by subtracting
dupLSAprocTime from totalSPFtime.
The accuracy of this test is crucially dependent on the amount of
time between the transmissions of the first and second LSAs. If too
much time elapsed, the test is meaningless because the SPF run will
complete before the second (duplicate) LSA is received. If the time
elapsed is less, then both LSAs will be handled before the SPF run is
scheduled and started, and thus the measurement would only be for the
handling of the duplicate LSA.
This test is also specified in [BLACKBOX].
Note: This test may not be accurate on systems that implement OSPF as
a multithreaded process, where the flooding takes place in a separate
process (or on a different processor) than shortest path first
computations.
It is also possible to measure the SPF time using white box tests
(using output supplied by the OSPF software implementer), such as the
following:
o Using reference topology 1 (Emulated Topology), establish a full
adjacency between the generator and the DUT.
o Inject a set of LSAs from the generator toward the DUT. Allow the
DUT to stabilize and install all best paths in the routing table,
etc.
o Change the link cost between the DUT and the generator (or the
link between the generator and the emulated network it is
advertising), such that a full SPF is required to run, although
only one piece of information is changed.
o Measure the amount of time required for the DUT to compute a new
shortest path tree as a result of the topology changes injected by
the generator. These measurements should be taken using available
show and debug information on the DUT.
Several caveats MUST be mentioned when a white box method of
measuring SPF time is used. For instance, such white box tests are
only applicable when testing various versions or variations within a
single implementation of the OSPF protocol. Further, the same set of
commands MUST be used in each iteration of such a test to ensure
consistent results.
There is an interesting relationship between the SPF times reported
by white box (internal) testing and black box (external) testing;
each of these two types of tests may be used as a "sanity check" on
the other by comparing results.
See [CONSIDERATIONS] for further discussion.
6. Basic Intra-area OSPF Tests
These tests measure the performance of an OSPF implementation for
basic intra-area tasks, including:
o Forming Adjacencies on Point-to-Point Link (Initialization)
o Forming Adjacencies on Point-to-Point Links
o Link Up with Information Already in the Database
o Initial convergence Time on a Designated Router Electing
(Broadcast) Network
o Link Down with Layer 2 Detection
o Link Down with Layer 3 Detection
o Designated Router Election Time on A Broadcast Network
6.1. Forming Adjacencies on Point-to-Point Link (Initialization)
This test measures the time required to form an OSPF adjacency from
the time a layer two (data link) connection is formed between two
devices running OSPF.
o Use reference topology 1 (Emulated Topology), beginning with the
link between the generator and DUT disabled on the DUT. OSPF
should be configured and operating on both devices.
o Inject a set of LSAs from the generator toward the DUT.
o Bring the link up at the DUT, noting the time when the link
carrier is established on the generator.
o Note the time when the acknowledgement for the last LSA
transmitted from the DUT is received on the generator.
The time between the carrier establishment and the acknowledgement
for the last LSA transmitted by the generator should be taken as the
total amount of time required for the OSPF process on the DUT to
react to a link up event with the set of LSAs injected, including the
time required for the operating system to notify the OSPF process
about the link up, etc. The acknowledgement for the last LSA
transmitted is used instead of the last acknowledgement received in
order to prevent timing skews due to retransmitted acknowledgements
or LSAs.
6.2. Forming Adjacencies on Point-to-Point Links
This test measures the time required to form an adjacency from the
time the first communication occurs between two devices running OSPF.
o Using reference topology 1 (Emulated Topology), configure the DUT
and the generator so that traffic can be passed along the link
between them.
o Configure the generator so that OSPF is running on the point-to-
point link toward the DUT, and inject a set of LSAs.
o Configure the DUT so that OSPF is initialized, but not running on
the point-to-point link between the DUT and the generator.
o Enable OSPF on the interface between the DUT and the generator on
the DUT.
o Note the time of the first hello received from the DUT on the
generator.
o Note the time of the acknowledgement from the DUT for the last LSA
transmitted on the generator.
The time between the first hello received and the acknowledgement for
the last LSA transmitted by the generator should be taken as the
total amount of time required for the OSPF process on the DUT to
build a FULL neighbor adjacency with the set of LSAs injected. The
acknowledgement for the last LSA transmitted is used instead of the
last acknowledgement received in order to prevent timing skews due to
retransmitted acknowledgements or LSAs.
6.3. Forming Adjacencies with Information Already in the Database
o Using reference topology 2 (Generator and Collector), configure
all three devices to run OSPF.
o Configure the DUT so that the link between the DUT and the
generator is disabled.
o Inject a set of LSAs into the network from the generator; the DUT
should receive these LSAs through normal flooding from the
collector.
o Enable the link between the DUT and the generator.
o Note the time of the first hello received from the DUT on the
generator.
o Note the time of the last DBD (Database Description) received on
the generator.
o Note the time of the acknowledgement from the DUT for the last LSA
transmitted on the generator.
The time between the hello received by the generator from the DUT and
the acknowledgement for the last LSA transmitted by the generator
should be taken as the total amount of time required for the OSPF
process on the DUT to build a FULL neighbor adjacency with the set of
LSAs injected. In this test, the DUT is already aware of the entire
network topology, so the time required should only include the
processing of DBDs exchanged when in EXCHANGE state, the time to
build a new router LSA containing the new connection information, and
the time required to flood and acknowledge this new router LSA.
The acknowledgement for the last LSA transmitted is used instead of
the last acknowledgement received in order to prevent timing skews
due to retransmitted acknowledgements or LSAs.
6.4. Designated Router Election Time on a Broadcast Network
o Using reference topology 3 (Broadcast Network), configure R1 to be
the designated router on the link, and the DUT to be the backup
designated router.
o Enable OSPF on the common broadcast link on all the routers in the
test bed.
o Disable the broadcast link on R1.
o Note the time of the last hello received from R1 on R2.
o Note the time of the first network LSA generated by the DUT as
received on R2.
The time between the last hello received on R2 and the first network
LSA generated by the DUT should be taken as the amount of time
required for the DUT to complete a designated router election
computation. Note that this test includes the dead interval timer at
the DUT, so this time may be factored out, or the hello and dead
intervals may be reduced to lessen these timers' impact on the
overall test times. All changed timers, the number of routers
connected to the link, and other variable factors should be noted in
the test results.
Note: If R1 sends a "goodbye hello", typically a hello with its
neighbor list empty, in the process of shutting down its interface,
using the time when this hello is received instead of the time when
the last one was would provide a more accurate measurement.
6.5. Initial Convergence Time on a Broadcast Network, Test 1
o Using reference topology 3 (Broadcast Network), begin with the DUT
connected to the network with OSPF enabled. OSPF should be
enabled on R1, but the broadcast link should be disabled.
o Enable the broadcast link between R1 and the DUT. Note the time
of the first hello received by R1.
o Note the time when the first network LSA is flooded by the DUT at
R1.
o The difference between the first hello and the first network LSA
is the time required by the DUT to converge on this new topology.
This test assumes that the DUT will be the designated router on the
broadcast link. A similar test could be designed to test the
convergence time when the DUT is not the designated router.
This test maybe performed with a varying number of devices attached
to the broadcast network, and with varying sets of LSAs being
advertised to the DUT from the routers attached to the broadcast
network. Variations in the LSA sets and other factors should be
noted in the test results.
The time required to elect a designated router, as measured in
Section 6.4, above, may be subtracted from the results of this test
to provide just the convergence time across a broadcast network.
Note that although all the other tests in this document include route
calculation time in the convergence time, as described in [TERM],
this test may not include route calculation time in the resulting
measured convergence time, because initial route calculation may
occur after the first network LSA is flooded.
6.6. Initial Convergence Time on a Broadcast Network, Test 2
o Using reference topology 3 (Broadcast Network), begin with the DUT
connected to the network with OSPF enabled. OSPF should be
enabled on R1, but the broadcast link should be disabled.
o Enable the broadcast link between R1 and the DUT. Note the time
of the first hello transmitted by the DUT with a designated router
listed.
o Note the time when the first network LSA is flooded by the DUT at
R1.
o The time difference between the first hello with a designated
router lists and the first network LSA is the period required by
the DUT to converge on this new topology.
6.7. Link Down with Layer 2 Detection
o Using reference topology 4 (Parallel Links), begin with OSPF in
the Full state between the generator and the DUT. Both links
should be point-to-point links with the ability to notify the
operating system immediately upon link failure.
o Disable link 1; this should be done in such a way that the
keepalive timers at the data link layer will have no impact on the
DUT recognizing the link failure (the operating system in the DUT
should recognize this link failure immediately). Disconnecting
the cable on the generator end would be one possibility; shutting
the link down would be another.
o Note the time of the link failure on the generator.
o At the generator, note the time of the receipt of the new router
LSA from the DUT notifying the generator of the link 2 failure.
The difference in the time between the initial link failure and
the receipt of the LSA on the generator across link 2 should be
taken as the time required for an OSPF implementation to recognize
and process a link failure, including the time required to
generate and flood an LSA describing the link down event to an
adjacent neighbor.
6.8. Link Down with Layer 3 Detection
o Using reference topology 4 (Parallel Links), begin with OSPF in
the Full state between the generator and the DUT.
o Disable OSPF processing on link 1 from the generator. This should
be done in such a way that it does not affect link status; the DUT
MUST note the failure of the adjacency through the dead interval.
o At the generator, note the time of the receipt of the new router
LSA from the DUT notifying the generator of the link 2 failure.
The difference in the time between the initial link failure and the
receipt of the LSA on the generator across link 2 should be taken as
the time required for an OSPF implementation to recognize and process
an adjacency failure.
7. Security Considerations
This document does not modify the underlying security considerations
in [OSPF].
8. Acknowledgements
Thanks to Howard Berkowitz (hcb@clark.net) for his encouragement and
support. Thanks also to Alex Zinin (zinin@psg.net), Gurpreet Singh
(Gurpreet.Singh@SpirentCom.com), and Yasuhiro Ohara
(yasu@sfc.wide.ad.jp) for their comments.
9. Normative References
[OSPF] Moy, J., "OSPF Version 2", STD 54, RFC 2328, April
1998.
[TERM] Manral, V., White, R., and A. Shaikh, "OSPF
Benchmarking Terminology and Concepts", RFC 4062,
April 2005.
[CONSIDERATIONS] Manral, V., White, R., and A. Shaikh,
"Considerations When Using Basic OSPF Convergence
Benchmarks", RFC 4063, April 2005.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
10. Informative References
[INTERCONNECT] Bradner, S. and J. McQuaid, "Benchmarking
Methodology for Network Interconnect Devices", RFC
2544, March 1999.
[FIB-TERM] Trotter, G., "Terminology for Forwarding Information
Base (FIB) based Router Performance", RFC 3222,
December 2001.
[BLACKBOX] Shaikh, A. and Greenberg, A., "Experience in Black-
box OSPF measurement", Proc. ACM SIGCOMM Internet
Measurement Workshop (IMW), November 2001
Authors' Addresses
Vishwas Manral
SiNett Corp,
Ground Floor,
Embassy Icon Annexe,
2/1, Infantry Road,
Bangalore, India
EMail: vishwas@sinett.com
Russ White
Cisco Systems, Inc.
7025 Kit Creek Rd.
Research Triangle Park, NC 27709
EMail: riw@cisco.com
Aman Shaikh
AT&T Labs (Research)
180 Park Av, PO Box 971
Florham Park, NJ 07932
EMail: ashaikh@research.att.com
Full Copyright Statement
Copyright (C) The Internet Society (2005).
This document is subject to the rights, licenses and restrictions
contained in BCP 78, and except as set forth therein, the authors
retain all their rights.
This document and the information contained herein are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Intellectual Property
The IETF takes no position regarding the validity or scope of any
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; nor does it represent that it has
made any independent effort to identify any such rights. Information
on the procedures with respect to rights in RFC documents can be
found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementers or users of this
specification can be obtained from the IETF on-line IPR repository at
http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights that may cover technology that may be required to implement
this standard. Please address the information to the IETF at ietf-
ipr@ietf.org.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.
ISO Plain Language
On this page
§1Why the Framework Matters
There is a growing interest in routing protocol convergence testing, with many people looking at various tests to determine how long it takes for a network to converge after various conditions occur. The major problem with this sort of testing is that the framework of the tests has a major impact on the results. Deciding when a network is considered converged, deciding what parts of the router's operation are inside the test, and other such choices have a major impact on the apparent performance that a routing protocol provides.
RFC 4061 attempts to provide a framework for OSPF performance testing, and to provide some tests for measuring some aspects of OSPF performance. The motivation is to give the user a set of tests that can produce comparable data from various vendors, with which to evaluate OSPF protocol performance on the devices.
The memo is careful about its own wording. The RFC 2119 key words are used to ensure methodological control, which the memo says is very important in the specification of benchmarks, and it says it does not specify a network-related protocol.
Timing a router is easy to do badly. If two people time the same change but one of them stops the clock at a different moment, the numbers mean nothing to each other. This memo fixes the clock rules so that the numbers can be compared.
§3What Is and Is Not Measured
Although the memo describes a specific set of tests aimed at characterizing the single router control plane convergence performance of OSPF processes in routers or other boxes that incorporate OSPF functionality, a key objective is to propose methodologies that produce directly comparable convergence-related measurements.
Two things are outside its scope. The first is the interaction of convergence and forwarding, because testing is restricted to events occurring within the control plane. Forwarding performance is the primary focus of RFC 2544, and it is expected to be dealt with in work that follows from RFC 3222. The second is inter-area route generation, AS-external route generation, and simultaneous traffic on the control and data paths within the device under test. Although the tests measure SPF time, flooding times and other aspects of OSPF convergence performance, the memo does not provide tests for measuring external or summary route generation, route translation, or other OSPF inter-area and external routing performance. Those areas are expected to be dealt with in a later document.
The memo also says the tests should be run more than once, since a single test run cannot be relied on to produce statistically sound results, and that the number of test runs and any variations between the tests should be recorded in the test results, with RFC 4062 saying what items should be recorded.
Everything here happens inside the routing process on one router. How fast the router forwards packets afterwards is a separate question, and so are routes that come from outside the area. The memo also insists that a test be repeated, because one run is not a measurement.
§4The Reference Topologies
Four reference topologies are used throughout the tests, and they are gathered into one section to avoid repetition.
Reference Topology 1 is the emulated topology, a simple back-to-back configuration in which the device under test is connected to a generator. The link between the generator and the device under test is assumed to be a point-to-point link, while the connections within the generator represent some emulated topology.
Reference Topology 2 is the generator and collector arrangement, in which a collector is connected to the device under test, the device under test is connected to a generator, and the generator is connected to an emulated topology. All routers are connected through point-to-point links, and the cost of all links is assumed to be the same unless otherwise noted.
Reference Topology 3 is a broadcast network, in which the device under test and a number of other routers, labelled R1 and R2 in the diagram, sit on a common broadcast link. Any number of routers could be included on the common broadcast network.
Reference Topology 4 uses parallel links, in which the device under test reaches the generator over two separate links, and the generator in turn reaches an emulated topology.
In all cases the tests and topologies are designed to allow performance measurements to be taken all on a single device, whether that is the device under test or some other device in the network. This eliminates the need for synchronized clocks within the test networks.
Four bench layouts are used, and they are drawn so that every measurement can be taken at one end. That matters, because if two machines had to be compared, their clocks would have to agree exactly.
1. Emulated topology
DUT----Generator----( emulated topology )
2. Generator and collector
( )
Collector-----DUT-----Generator--( emulated topology )
\ / ( )
\------------/
3. Broadcast network
DUT R1 R2
| | |
-+------+------+-----.....
4. Parallel links
/--(link 1)-----\ ( )
DUT Generator--( emulated topology )
\--(link 2)-----/ ( )§5The Three Basic Measurements
The basic tests measure three aspects of the OSPF implementation as a process on the device under test: the time required to process an LSA, the flooding time, and the Shortest Path First computation.
Time required to process an LSA. Using the emulated topology, the test begins with all links up and a full adjacency established between the device under test and the generator. The memo warns that the generator has no direct knowledge of the state of the adjacency on the device under test: the adjacency may be in the Full state on the generator and the device under test may still be requesting LSAs from it, and that processing will affect the results. The generator should either wait until it sees the device under test's router-LSA listing the adjacency with the generator, or introduce a configurable delay before starting the test. The test then sends an LSA that is already in the device under test, a duplicate LSA, and notes the time between sending it and receiving the acknowledgement. That is the time taken to propagate the LSA and the acknowledgement, plus the processing time of the duplicate LSA, and it is called dupLSAprocTime. The test then sends a new LSA from the generator to the device under test, followed immediately by a duplicate LSA, meaning one that already resides in the database of the device under test but is not the same as the one just sent. The device under test acknowledges the second LSA immediately, and the time of that acknowledgement is newLSAprocTime. The time required to process the new LSA is computed by subtracting dupLSAprocTime from newLSAprocTime. The duplicate LSA cannot be the same as the one just sent, because of the MinLSInterval restriction. The memo notes that this time may or may not include the time required for flooding-related operations, depending on whether the implementation sends the acknowledgement before it floods the LSA further, after it does, or somewhere in between, so the measurement may not mean the same thing in all implementations.
Flooding time. Using the generator and collector topology, OSPF is enabled on all links and the devices are allowed to build full adjacencies. The collector is configured to block all flooding toward the device under test, while continuing to receive advertisements from it. A new set of LSAs is injected from the generator toward the collector and the device under test. On the collector, the memo asks for the time the flooding is complete across the link to the generator, and the time it is complete across the link from the device under test. The time from when the last LSA is received on the collector from the generator to when the last LSA is received on the collector from the device under test is the measurement. The memo explains why it matters: in link state protocols the loop-free nature of the network relies on the speed at which revised topology information is flooded. Depending on the number of LSAs flooded, their sizes, the number of LSUs and the rate of flooding, these numbers could vary by some amount, and the settings and variances should be reported with the test results.
SPF computation time. The test again uses the emulated topology, beginning with the device under test and the generator fully adjacent, and the default SPF timer on the device under test set to 0, so that any new LSA arriving immediately results in an SPF calculation. The generator injects a set of LSAs toward the device under test, which is allowed to converge and install all best paths in the local routing table. The duplicate LSA is sent again to establish dupLSAprocTime. The link cost between the generator and the emulated network it is advertising is then changed, and the new LSA is transmitted to the device under test, immediately followed by another LSA that duplicates one the generator has injected before, preferably a stub network somewhere within the emulated network. The generator has to make sure that outbound LSA packing is not performed for the duplicate LSAs, and that they are always sent in a separate Link-state Update packet, because if the LSA carrying the topology change and the duplicate LSA are in the same packet, the SPF starts after the duplicate LSA is acknowledged. The time between transmitting the second, duplicate, LSA and the acknowledgement for that LSA is totalSPFtime, and the total time required to run SPF is computed by subtracting dupLSAprocTime from it.
The accuracy of that test depends crucially on the time between the transmissions of the first and second LSAs. If too much time elapses, the test is meaningless, because the SPF run will complete before the second LSA is received. If less time elapses, both LSAs are handled before the SPF run is scheduled and started, and the measurement would only be for the handling of the duplicate LSA. The memo also notes that the test may not be accurate on systems that implement OSPF as a multithreaded process, where flooding takes place in a separate process, or on a different processor, from the shortest path first computations.
It is also possible to measure the SPF time with white box tests, using output supplied by the OSPF software implementer: establish a full adjacency, inject a set of LSAs and allow the device under test to stabilize and install all best paths, then change the link cost so that a full SPF is required to run although only one piece of information has changed, and measure the time required to compute a new shortest path tree, using the available show and debug information. Several caveats MUST be mentioned when this method is used. Such tests are only applicable when testing various versions or variations within a single implementation, and the same set of commands MUST be used in each iteration to ensure consistent results. The memo notes that the SPF times reported by white box and black box testing are related, and that each can be used as a sanity check on the other by comparing results, with further discussion in RFC 4063.
The three measurements are: how long the router takes to absorb a new routing update, how long that update takes to reach everyone else, and how long the router takes to work out new paths. The trick throughout is to send a second, known update straight afterwards, so that the time to acknowledge it gives a baseline that can be subtracted from the total.
- dupLSAprocTimeThe time between sending a duplicate LSA and receiving its acknowledgement. It covers propagation, the acknowledgement and the processing of the duplicate, and it is the baseline subtracted from the other measurements.
- newLSAprocTimeThe time of the acknowledgement of a second LSA, sent immediately after a new one. Subtracting dupLSAprocTime from it gives the time to process the new LSA.
- totalSPFtimeThe time between transmitting a duplicate LSA and receiving its acknowledgement, after a link cost change. Subtracting dupLSAprocTime from it gives the time to run SPF.
- Flooding timeThe time from when the last LSA reaches the collector from the generator to when the last LSA reaches the collector from the device under test.
§6The Intra-Area Tests
The intra-area tests measure the performance of an OSPF implementation for basic tasks inside an area: forming adjacencies on a point-to-point link during initialization, forming adjacencies on point-to-point links, forming adjacencies with information already in the database, initial convergence time on a designated router electing broadcast network, link down with layer two detection, link down with layer three detection, and designated router election time on a broadcast network.
Forming adjacencies on a point-to-point link (initialization) measures the time required to form an OSPF adjacency from the time a layer two, that is data link, connection is formed between two devices running OSPF. Using the emulated topology, the test begins with the link between the generator and the device under test disabled on the device under test, with OSPF configured and operating on both devices. A set of LSAs is injected from the generator toward the device under test, the link is brought up at the device under test while the time of link carrier establishment is noted on the generator, and the time of the acknowledgement for the last LSA transmitted from the device under test is noted on the generator. The time between the carrier establishment and that acknowledgement is the total time required for the OSPF process to react to a link up event with the set of LSAs injected, including the time for the operating system to notify the OSPF process. The acknowledgement for the last LSA transmitted is used rather than the last acknowledgement received, in order to prevent timing skews from retransmitted acknowledgements or LSAs.
Forming adjacencies on point-to-point links measures the time from the first communication between two devices running OSPF. The device under test is configured so that OSPF is initialized but not running on the point-to-point link toward the generator, and OSPF is then enabled on that interface. The time of the first hello received from the device under test on the generator is noted, along with the time of the acknowledgement from the device under test for the last LSA transmitted on the generator. The time between the two is the total time required to build a FULL neighbour adjacency with the set of LSAs injected.
Forming adjacencies with information already in the database uses the generator and collector topology, with all three devices running OSPF and the link between the device under test and the generator disabled. A set of LSAs is injected into the network from the generator, and the device under test receives them through normal flooding from the collector. The link is then enabled, and the times of the first hello received from the device under test, the last DBD, that is Database Description, received on the generator, and the acknowledgement from the device under test for the last LSA transmitted are noted. The time between the hello and the acknowledgement is the total time required to build a FULL neighbour adjacency. In this test the device under test is already aware of the entire network topology, so the time measured should only include the processing of DBDs exchanged while in EXCHANGE state, the time to build a new router LSA containing the new connection information, and the time to flood and acknowledge that new router LSA.
Designated router election time on a broadcast network uses the broadcast network topology, with R1 configured as the designated router on the link and the device under test as the backup designated router. OSPF is enabled on the common broadcast link on all routers, and then the broadcast link on R1 is disabled. The time of the last hello received from R1 on R2 is noted, along with the time of the first network LSA generated by the device under test as received on R2. The time between them is the amount of time required for the device under test to complete a designated router election computation. This test includes the dead interval timer at the device under test, so the time may be factored out, or the hello and dead intervals may be reduced to lessen their impact. All changed timers, the number of routers connected to the link and other variable factors should be noted in the test results. If R1 sends a goodbye hello, typically a hello with its neighbour list empty, as it shuts its interface down, using the time when that hello is received gives a more accurate measurement.
Initial convergence time on a broadcast network, test 1 uses the broadcast network topology, beginning with the device under test connected with OSPF enabled and OSPF enabled on R1 but the broadcast link disabled. The broadcast link between R1 and the device under test is enabled, the time of the first hello received by R1 is noted, and the time when the first network LSA is flooded by the device under test at R1 is noted. The difference between the two is the time required by the device under test to converge on this new topology. The test assumes the device under test will be the designated router, and a similar test could be designed for the case where it is not. The time required to elect a designated router, as measured in the previous test, may be subtracted to give just the convergence time across a broadcast network. Unlike all the other tests, this one may not include route calculation time, because the initial route calculation may occur after the first network LSA is flooded. Test 2 uses the same setup, but takes the time of the first hello transmitted by the device under test with a designated router listed, rather than the hello received on R1, and the difference between that and the first network LSA is the period required by the device under test to converge on the new topology.
Link down with layer two detection uses the parallel links topology, beginning with OSPF in the Full state between the generator and the device under test, with both links point-to-point and able to notify the operating system immediately on link failure. Link 1 is disabled in such a way that the keepalive timers at the data link layer have no impact on the device under test recognizing the failure, for example by disconnecting the cable at the generator end or shutting the link down. The time of the link failure on the generator is noted, and then the time of receipt on the generator of the new router LSA from the device under test notifying it of the link 2 failure. The difference is the time required for an OSPF implementation to recognize and process a link failure, including the time to generate and flood an LSA describing the link down event to an adjacent neighbour. Link down with layer three detection uses the same topology, but instead disables OSPF processing on link 1 from the generator, in such a way that it does not affect link status, so the device under test MUST note the failure of the adjacency through the dead interval. The same two timestamps are then taken, and the difference is the time required to recognize and process an adjacency failure.
These are the day-to-day events: a link comes up, a link goes down, a neighbour is elected on a shared network. Each test says exactly which two timestamps to take and what the gap between them means. The last pair is the interesting one: a cable pulled is noticed at once, while a silent neighbour is only noticed when its dead interval expires.
| Aspect | Layer two detection | Layer three detection |
|---|---|---|
| How the failure is seen | The operating system notifies the router immediately | The adjacency is declared dead when the dead interval expires |
| How the test is set up | Disable link 1 so that the data link keepalive timers have no impact | Disable OSPF processing on link 1, without affecting link status |
| What is measured | The time to recognise and process a link failure, and to flood the LSA describing it | The time to recognise and process an adjacency failure |
§7Security Considerations
The memo states in one line that it does not modify the underlying security considerations in RFC 2328.
Nothing here changes how OSPF authenticates itself, so the rules in RFC 2328 still apply.