Network Working Group G. Sadasivan

Network Working Group G. Sadasivan
Request for Comments: 5470 Rohati Systems
Category: Informational N. Brownlee
CAIDA | The University of Auckland
B. Claise
Cisco Systems, Inc.
J. Quittek
NEC
March 2009

Architecture for IP Flow Information Export

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) 2009 IETF Trust and the persons identified as the
document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents in effect on the date of
publication of this document (http://trustee.ietf.org/license-info).
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document.

This document may contain material from IETF Documents or IETF
Contributions published or made publicly available before November
10, 2008. The person(s) controlling the copyright in some of this
material may not have granted the IETF Trust the right to allow
modifications of such material outside the IETF Standards Process.
Without obtaining an adequate license from the person(s) controlling
the copyright in such materials, this document may not be modified
outside the IETF Standards Process, and derivative works of it may
not be created outside the IETF Standards Process, except to format
it for publication as an RFC or to translate it into languages other
than English.

Abstract

This memo defines the IP Flow Information eXport (IPFIX) architecture
for the selective monitoring of IP Flows, and for the export of
measured IP Flow information from an IPFIX Device to a Collector.

Sadasivan, et al. Informational [Page 1]

RFC 5470 IPFIX Architecture March 2009

Table of Contents

1. Introduction ....................................................3
1.1. Document Scope .............................................3
1.2. IPFIX Documents Overview ...................................3
2. Terminology .....................................................4
3. Examples of Flows ...............................................8
4. IPFIX Reference Model ..........................................10
5. IPFIX Functional and Logical Blocks ............................12
5.1. Metering Process ..........................................12
5.1.1. Flow Expiration ....................................12
5.1.2. Flow Export ........................................13
5.2. Observation Point .........................................13
5.3. Selection Criteria for Packets ............................13
5.3.1. Sampling Functions, Si .............................14
5.3.2. Filter Functions, Fi ...............................15
5.4. Observation Domain ........................................15
5.5. Exporting Process .........................................15
5.6. Collecting Process ........................................16
5.7. Summary ...................................................17
6. Overview of the IPFIX Protocol .................................18
6.1. Information Model Overview ................................19
6.2. Flow Records ..............................................19
6.3. Control Information .......................................20
6.4. Reporting Responsibilities ................................21
7. IPFIX Protocol Details .........................................21
7.1. The IPFIX Basis Protocol ..................................21
7.2. IPFIX Protocol on the Collecting Process ..................22
7.3. Support for Applications ..................................22
8. Export Models ..................................................23
8.1. Export with Reliable Control Connection ...................23
8.2. Collector Failure Detection and Recovery ..................23
8.3. Collector Redundancy ......................................24
9. IPFIX Flow Collection in Special Situations ....................24
10. Security Considerations .......................................25
10.1. Data Security ............................................25
10.1.1. Host-Based Security ...............................26
10.1.2. Authentication-Only ...............................26
10.1.3. Encryption ........................................26
10.2. IPFIX End-Point Authentication ...........................27
10.3. IPFIX Overload ...........................................27
10.3.1. Denial-of-Service (DoS) Attack Prevention .........27
11. IANA Considerations ...........................................28
11.1. Numbers Used in the Protocol .............................28
11.2. Numbers Used in the Information Model ....................29
12. Acknowledgements ..............................................29

Sadasivan, et al. Informational [Page 2]

RFC 5470 IPFIX Architecture March 2009

13. References ....................................................30
13.1. Normative References .....................................30
13.2. Informative References ...................................30

1. Introduction

There are several applications, e.g., usage-based accounting, traffic
profiling, traffic engineering, attack/intrusion detection, quality-
of-service (QoS) monitoring, that require Flow-based IP traffic
measurements. It is therefore important to have a standard way of
exporting information related to IP Flows. This document defines an
architecture for IP traffic Flow monitoring, measuring, and
exporting. It provides a high-level description of an IPFIX Device's
key components and their functions.

1.1. Document Scope

This document defines the architecture for IPFIX. Its main
objectives are to:

o Describe the key IPFIX architectural components, consisting of (at
least) IPFIX Devices and Collectors communicating using the IPFIX
protocol.

o Define the IPFIX architectural requirements, e.g., recovery,
security, etc.

o Describe the characteristics of the IPFIX protocol.

1.2. IPFIX Documents Overview

The IPFIX protocol provides network administrators with access to IP
Flow information. This document specifies the architecture for the
export of measured IP Flow information from an IPFIX Exporting
Process to a Collecting Process, per the requirements defined in RFC
3917 [1]. The IPFIX protocol document, RFC 5101 [3], specifies how
IPFIX data records and templates are carried via a congestion-aware
transport protocol, from IPFIX Exporting Process to IPFIX Collecting
Process. IPFIX has a formal description of IPFIX information
elements (fields), their name, type, and additional semantic
information, as specified in RFC 5102 [2]. Finally, RFC 5472 [4]
describes what type of applications can use the IPFIX protocol and
how they can use the information provided. Furthermore, it shows how
the IPFIX framework relates to other architectures and frameworks.

Note that the IPFIX system does not provide for remote configuration
of an IPFIX device. Instead, implementors must provide an effective
way to configure their IPFIX devices.

Sadasivan, et al. Informational [Page 3]

RFC 5470 IPFIX Architecture March 2009

2. Terminology

The definitions of basic IPFIX terms such as IP Traffic Flow,
Exporting Process, Collecting Process, Observation Point, etc., are
semantically identical with those found in the IPFIX requirements
document, RFC 3917 [1]. Some of the terms have been expanded for
more clarity when defining the protocol. Additional definitions
required for the architecture have also been defined. For terms that
are defined here and in RFC 5101 [3], the definitions are equivalent
in both documents.

* Observation Point

An Observation Point is a location in the network where IP packets
can be observed. Examples include: a line to which a probe is
attached, a shared medium, such as an Ethernet-based LAN, a single
port of a router, or a set of interfaces (physical or logical) of
a router.

Note that every Observation Point is associated with an
Observation Domain (defined below), and that one Observation Point
may be a superset of several other Observation Points. For
example, one Observation Point can be an entire line card. That
would be the superset of the individual Observation Points at the
line card's interfaces.

* Observation Domain

An Observation Domain is the largest set of Observation Points for
which Flow information can be aggregated by a Metering Process.
For example, a router line card may be an Observation Domain if it
is composed of several interfaces, each of which is an Observation
Point. In the IPFIX Message it generates, the Observation Domain
includes its Observation Domain ID, which is unique per Exporting
Process. That way, the Collecting Process can identify the
specific Observation Domain from the Exporter that sends the IPFIX
Messages. Every Observation Point is associated with an
Observation Domain. It is recommended that Observation Domain IDs
also be unique per IPFIX Device.

* IP Traffic Flow or Flow

There are several definitions of the term 'flow' being used by the
Internet community. Within the context of IPFIX we use the
following definition:

Sadasivan, et al. Informational [Page 4]

RFC 5470 IPFIX Architecture March 2009

A Flow is defined as a set of IP packets passing an Observation
Point in the network during a certain time interval. All packets
belonging to a particular Flow have a set of common properties.
Each property is defined as the result of applying a function to
the values of:

1. one or more packet header fields (e.g., destination IP
address), transport header fields (e.g., destination port
number), or application header fields (e.g., RTP header fields
[5]).

2. one or more characteristics of the packet itself (e.g., number
of MPLS labels)

3. one or more fields derived from packet treatment (e.g., next
hop IP address, output interface)

A packet is defined as belonging to a Flow if it completely
satisfies all the defined properties of the Flow.

This definition covers the range from a Flow containing all
packets observed at a network interface to a Flow consisting of
just a single packet between two applications. It includes
packets selected by a sampling mechanism.

* Flow Key

Each of the fields that:

1. belongs to the packet header (e.g., destination IP address),

2. is a property of the packet itself (e.g., packet length),

3. is derived from packet treatment (e.g., Autonomous System (AS)
number), and

4. is used to define a Flow

is termed a Flow Key.

* Flow Record

A Flow Record contains information about a specific Flow that was
observed at an Observation Point. A Flow Record contains measured
properties of the Flow (e.g., the total number of bytes for all
the Flow's packets) and usually characteristic properties of the
Flow (e.g., source IP address).

Sadasivan, et al. Informational [Page 5]

RFC 5470 IPFIX Architecture March 2009

* Metering Process

The Metering Process generates Flow Records. Inputs to the
process are packet headers and characteristics observed at an
Observation Point, and packet treatment at the Observation Point
(for example, the selected output interface).

The Metering Process consists of a set of functions that includes
packet header capturing, timestamping, sampling, classifying, and
maintaining Flow Records.

The maintenance of Flow Records may include creating new records,
updating existing ones, computing Flow statistics, deriving
further Flow properties, detecting Flow expiration, passing Flow
Records to the Exporting Process, and deleting Flow Records.

* Exporting Process

The Exporting Process sends Flow Records to one or more Collecting
Processes. The Flow Records are generated by one or more Metering
Processes.

* Exporter

A device that hosts one or more Exporting Processes is termed an
Exporter.

* IPFIX Device

An IPFIX Device hosts at least one Exporting Process. It may host
further Exporting Processes and arbitrary numbers of Observation
Points and Metering Processes.

* Collecting Process

A Collecting Process receives Flow Records from one or more
Exporting Processes. The Collecting Process might process or
store received Flow Records, but such actions are out of scope for
this document.

* Collector

A device that hosts one or more Collecting Processes is termed a
Collector.

Sadasivan, et al. Informational [Page 6]

RFC 5470 IPFIX Architecture March 2009

* Template

A Template is an ordered sequence of <type, length> pairs used to
completely specify the structure and semantics of a particular set
of information that needs to be communicated from an IPFIX Device
to a Collector. Each Template is uniquely identifiable by means
of a Template ID.

* Control Information, Data Stream

The information that needs to be exported from the IPFIX Device
can be classified into the following categories:

Control Information

This includes the Flow definition, selection criteria for
packets within the Flow sent by the Exporting Process, and
templates describing the data to be exported. Control
Information carries all the information needed for the end-
points to understand the IPFIX protocol, and specifically for
the Collector to understand and interpret the data sent by the
sending Exporter.

Data Stream

This includes Flow Records carrying the field values for the
various observed Flows at each of the Observation Points.

* IPFIX Message

An IPFIX Message is a message originating at the Exporting Process
that carries the IPFIX records of this Exporting Process and whose
destination is a Collecting Process. An IPFIX Message is
encapsulated at the transport layer.

* Information Element

An Information Element is a protocol and encoding-independent
description of an attribute that may appear in an IPFIX Record.
The IPFIX information model, RFC 5102 [2], defines the base set of
Information Elements for IPFIX. The type associated with an
Information Element indicates constraints on what it may contain
and also determines the valid encoding mechanisms for use in
IPFIX.

Sadasivan, et al. Informational [Page 7]

RFC 5470 IPFIX Architecture March 2009

3. Examples of Flows

Some examples of Flows are listed below. In the IPv4 examples, we
use interface addresses in three different 26-bit (/26) subnets. In
the IPv6 examples, we use 'mac addr-nn' in the low-order 64 bits to
indicate the IEEE MAC (Media Access Control) address of host
interface nn.

Example 1: Flow Keys define the different fields by which Flows are
distinguished. The different combination of their field values
creates unique Flows. If {source IP address, destination IP address,
DSCP} are Flow Keys, then all of these are different Flows:

1. {192.0.2.1, 192.0.2.65, 4}
2. {192.0.2.23, 192.0.2.67, 4}
3. {192.0.2.23, 192.0.2.67, 2}
4. {192.0.2.129, 192.0.2.67, 4}

5. {2001:DB8::0:mac-addr-01, 2001:DB8::1:mac-addr-11, 4}
6. {2001:DB8::0:mac-addr-02, 2001:DB8::1:mac-addr-13, 4}
7. {2001:DB8::0:mac-addr-02, 2001:DB8::1:mac-addr-13, 2}
8. {2001:DB8::2:mac-addr-21, 2001:DB8::1:mac-addr-13, 4}

Example 2: A mask function can be applied to all the packets that
pass through an Observation Point, in order to aggregate some values.
This could be done by defining the set of Flow Keys as {source IP
address, destination IP address, DSCP} as in Example 1 above, and
applying functions that mask out the source and destination IP
addresses (least significant 6 bits for IPv4, 64 bits for IPv6). The
eight Flows from Example 1 would now be aggregated into six Flows by
merging the Flows 1+2 and 5+6 into single Flows:

1. {192.0.2.0/26, 192.0.2.64/26, 4}
2. {192.0.2.0/26, 192.0.2.64/26, 2}
3. {192.0.2.128/26, 192.0.2.64/26, 4}

4. {2001:DB8::0/64, 2001:DB8::1/64, 4}
5. {2001:DB8::0/64, 2001:DB8::1/64, 2}
6. {2001:DB8::2/64, 2001:DB8::1/64, 4}

Example 3: A filter defined by some Flow Key values can be applied on
all packets that pass the Observation Point, in order to select only
certain Flows. The filter is defined by choosing fixed values for
specific Keys from the packet.

All the packets that go from a customer network 192.0.2.0/26 to
another customer network 192.0.2.64/26 with DSCP value of 4 define a
Flow. All other combinations don't define a Flow and are not taken

Sadasivan, et al. Informational [Page 8]

RFC 5470 IPFIX Architecture March 2009

into account. The three Flows from Example 2 would now be reduced to
one Flow by filtering out Flows 2 and 3, leaving only Flow 1,
{192.0.2.0/26, 192.0.2.64/26, 4}.

Similarly, for the IPv6 packets in the examples above, one could
filter out Flows 5 and 6 to leave Flow 4.

The above examples can be thought of as a function F() taking as
input {source IP address, destination IP address, DSCP}. The
function selects only the packets that satisfy all three of the
following conditions:

1. Mask out the least significant 6 bits of source IP address, match
against 192.0.2.0.

2. Mask out the least significant 6 bits of destination IP address,
match against 192.0.2.64.

3. Only accept DSCP value equal to 4.

Depending on the values of {source IP address, destination IP
address, DSCP} of the different observed packets, the Metering
Process function F() would choose/filter/aggregate different sets of
packets, which would create different Flows. For example, for
various combinations of values of {source IP address, destination IP
address, DSCP}, F(source IP address, destination IP address, DSCP)
would result in the definition of one or more Flows.

Sadasivan, et al. Informational [Page 9]

RFC 5470 IPFIX Architecture March 2009

4. IPFIX Reference Model

The figure below shows the reference model for IPFIX. This figure
covers the various possible scenarios that can exist in an IPFIX
system.

+----------------+ +----------------+
|[*Application 1]| ... |[*Application n]|
+--------+-------+ +-------+--------+
^ ^
| |
+ = = = = -+- = = = = +
^
|
+------------------------+ +-------+------------------+
|IPFIX Exporter | | Collector(1) |
|[Exporting Process(es)] |<---------->| [Collecting Process(es)] |
+------------------------+ +--------------------------+
.... ....
+------------------------+ +---------------------------+
|IPFIX Device(i) | | Collector(j) |
|[Observation Point(s)] |<--------->| [Collecting Process(es)] |
|[Metering Process(es)] | +---->| [*Application(s)] |
|[Exporting Process(es)] | | +---------------------------+
+------------------------+ .
.... . ....
+------------------------+ | +--------------------------+
|IPFIX Device(m) | | | Collector(n) |
|[Observation Point(s)] |<----+---->| [Collecting Process(es)] |
|[Metering Process(es)] | | [*Application(s)] |
|[Exporting Process(es)] | +--------------------------+
+------------------------+

The various functional components are indicated within brackets [].
The functional components within [*] are not part of the IPFIX
architecture. The interfaces shown by "<----->" are defined by the
IPFIX architecture, but those shown by "<= = = =>" are not.

Figure 1: IPFIX Reference Model

Sadasivan, et al. Informational [Page 10]

RFC 5470 IPFIX Architecture March 2009

The figure below shows a typical IPFIX Device where the IPFIX
components are shown in rectangular boxes.

+--------------------------------------------------+
| IPFIX Device |
| +-----+ |
| +------- ... ------------+---------> | |
| | | | | |
| +----+----+ +----+----+ | | |
| |Metering | |Metering | | E | |
| |Process 1| |Process N| | x | |
| +---------+ +---------+ | p | |
| ^ ^ | o | |
| +------+--------+ +---------+------+ | r | |
| | Obsv Domain 1 | | Obsv Domain N | | t | |
| |+-----+-------+| |+-------+------+| | i | |
| ||Obsv Pt 1..j || ... ||Obsv Pt j+1..M|| | n | |
| |+-------------+| |+--------------+| | g | | Export
Packets | +------^--------+ +---------^------+ | | | packets
--->----+--------+---------- ... ---------+ | | | to
In | | +--------->
| . . . . . | | |Collector
| | | |
| +------ ... -------------+---------> | |
| | | | | |
| +----+----+ +----+----+ | P | |
| |Metering | |Metering | | r | |
| |Process 1| |Process N| | o | |
| +---------+ +---------+ | c | |
| ^ ^ | e | |
| +------+--------+ +---------+------+ | s | |
| | Obsv Domain 1 | | Obsv Domain N | | s | |
| |+-----+-------+| |+-------+------+| | | |
| ||Obsv Pt 1..k || ... ||Obsv Pt k+1..M|| | | |
| |+-------------+| |+--------------+| | | |
Packets | +------^--------+ +---------^------+ +-----+ |
--->----+--------+---------- ... ---------+ |
In | |
+--------------------------------------------------+

Figure 2: IPFIX Device

Sadasivan, et al. Informational [Page 11]

RFC 5470 IPFIX Architecture March 2009

5. IPFIX Functional and Logical Blocks

5.1. Metering Process

Every Observation Point in an IPFIX Device, participating in Flow
measurements, must be associated with at least one Metering Process.
Every packet coming into an Observation Point goes into each of the
Metering Processes associated with the Observation Point. Broadly,
each Metering Process observes the packets that pass an Observation
Point, does timestamping, and classifies the packets into Flow(s)
based on the selection criteria.

The Metering Process is a functional block that manages all the Flows
generated from an Observation Domain. The typical functions of a
Metering Process may include:

o Maintaining database(s) of all the Flow Records from an
Observation Domain. This includes creating new Flow Records,
updating existing ones, computing Flow Records statistics,
deriving further Flow properties, and adding non-Flow-specific
information based on the packet treatment (in some cases, fields
like AS numbers, router state, etc.)

o Maintaining statistics about the Metering Process itself, such as
Flow Records generated, packets observed, etc.

5.1.1. Flow Expiration

A Flow is considered to have expired under the following conditions:

1. If no packets belonging to the Flow have been observed for a
certain period of time. This time period should be configurable
at the Metering Process, with a minimum value of 0 seconds for
immediate expiration. Note that a zero timeout would report a
Flow as a sequence of single-packet Flows.

2. If the IPFIX Device experiences resource constraints, a Flow may
be prematurely expired (e.g., lack of memory to store Flow
Records).

3. For long-running Flows, the Metering Process should expire the
Flow on a regular basis or based on some expiration policy. This
periodicity or expiration policy should be configurable at the
Metering Process. When a long-running Flow is expired, its Flow
Record may still be maintained by the Metering Process so that
the Metering Process does not need to create a new Flow Record
for further observed packets of the same Flow.

Sadasivan, et al. Informational [Page 12]

RFC 5470 IPFIX Architecture March 2009

5.1.2. Flow Export

The Exporting Process decides when and whether to export an expired
Flow. A Flow can be exported because it expired for any of the
reasons mentioned in Section 5.1.1, "Flow Expiration". For example:
the Exporting Process exports a portion of the expired Flows every
'x' seconds.

For long-lasting Flows, the Exporting Process should export the Flow
Records on a regular basis or based on some export policy. This
periodicity or export policy should be configurable at the Exporting
Process.

5.2. Observation Point

A Flow Record can be better analysed if the Observation Point from
which it was measured is known. As such, it is recommended that
IPFIX Devices send this information to Collectors. In cases where
there is a single Observation Point or where the Observation Point
information is not relevant, the Metering Process may choose not to
add the Observation Point information to the Flow Records.

5.3. Selection Criteria for Packets

A Metering Process may define rules so that only certain packets
within an incoming stream of packets are chosen for measurement at an
Observation Point. This may be done by one of the two methods
defined below or a combination of them, in either order. A
combination of each of these methods can be adopted to select the
packets, i.e., one can define a set of methods {F1, S1, F2, S2, S3}
executed in a specified sequence at an Observation Point to select
particular Flows.

The figure below shows the operations that may be applied as part of
a typical Metering Process.

Sadasivan, et al. Informational [Page 13]

RFC 5470 IPFIX Architecture March 2009

+---------------------------+
| packet header capturing |
+---------------------------+
|
v
+---------------------------+
| timestamping |
+---------------------------+
|
v
+---------------> +
| |
| v
| +----------------------------------------------+
| | sampling Si (1:1 in case of no sampling) |
| +----------------------------------------------+
| |
| v
| +----------------------------------------------+
| | filtering Fi (select all when no criteria) |
| +----------------------------------------------+
| |
| v
+-----------------+
|
v
+---------------------------+
| Flows |
+---------------------------+

Figure 3: Selection Criteria for Packets

Note that packets could be selected before or after any IP
processing, i.e., before there is any IP checksum validation, IP
filtering, etc., or after one or more of these steps. This has an
impact on what kinds of traffic (or erroneous conditions) IPFIX can
observe. It is recommended that packets are selected after their
checksums have been verified.

5.3.1. Sampling Functions, Si

A sampling function determines which packets within a stream of
incoming packets are selected for measurement, i.e., packets that
satisfy the sampling criteria for this Metering Process.

Example: sample every 100th packet that was received at an
Observation Point.

Sadasivan, et al. Informational [Page 14]

RFC 5470 IPFIX Architecture March 2009

Choosing all the packets is a special case where the sampling rate is
1:1.

5.3.2. Filter Functions, Fi

A Filter Function selects only those incoming packets that satisfy a
function on fields defined by the packet header fields, fields
obtained while doing the packet processing, or properties of the
packet itself.

Example: Mask/Match of the fields that define a filter. A filter
might be defined as {Protocol == TCP, Destination Port < 1024}.

Several such filters could be used in any sequence to select packets.
Note that packets selected by a (sequence of) filter functions may be
further classified by other filter functions, i.e., the selected
packets may belong to several Flows, any or all of which are
exported.

5.4. Observation Domain

The Observation Domain is a logical block that presents a single
identity for a group of Observation Points within an IPFIX Device.
Each {Observation Point, Metering Process} pair belongs to a single
Observation Domain. An IPFIX Device could have multiple Observation
Domains, each of which has a subset of the total set of Observation
Points in it. Each Observation Domain must carry a unique ID within
the context of an IPFIX Device. Note that in the case of multiple
Observation Domains, a unique ID per Observation Domain must be
transmitted as a parameter to the Exporting Function. That unique ID
is referred to as the IPFIX Observation Domain ID.

5.5. Exporting Process

The Exporting Process is the functional block that sends data to one
or more IPFIX Collectors using the IPFIX protocol. On one side, the
Exporting Process interfaces with Metering Process(es) to get Flow
Records; while on the other side, it talks to a Collecting Process on
the Collector(s).

There may be additional rules defined within an Observation Domain so
that only certain Flow Records are exported. This may be done by
either one or a combination of Si and Fi, as described in
Section 5.3, "Selection Criteria for Packets".

Example: Only the Flow Records that meet the following selection
criteria are exported:

Sadasivan, et al. Informational [Page 15]

RFC 5470 IPFIX Architecture March 2009

1. All Flow Records whose destination IP address matches
{192.0.33.5}.

2. Every other (i.e., sampling rate 1 in 2) Flow Record whose
destination IP address matches {192.0.11.30}.

5.6. Collecting Process

Collecting Processes use a Flow Record's Template ID to interpret
that Flow Record's Information Elements. To allow this, an IPFIX
Exporter must ensure that an IPFIX Collector knows the Template ID
for each incoming Flow Record. To interpret incoming Flow Records,
an IPFIX Collector may also need to know the function F() that was
used by the Metering Process for each Flow.

The functions of the Collecting Process must include:

o Identifying, accepting, and decoding the IPFIX Messages from
different <Exporting Process, Observation Domain> pairs.

o Storing the Control Information and Flow Records received from an
IPFIX Device.

At a high level, the Collecting Process:

1. Receives and stores the Control Information.

2. Decodes and stores the Flow Records using the Control
Information.

Sadasivan, et al. Informational [Page 16]

RFC 5470 IPFIX Architecture March 2009

5.7. Summary

The figure below shows the functions performed in sequence by the
various functional blocks in an IPFIX Device.

Packet(s) coming into Observation Point(s)
| |
v v
+----------------+-------------------------+ +-----+-------+
| Metering Process on an | | |
| Observation Point | | |
| | | |
| packet header capturing | | |
| | |...| Metering |
| timestamping | | Process N |
| | | | |
| +----->+ | | |
| | | | | |
| | sampling Si (1:1 in case of no | | |
| | | sampling) | | |
| | filtering Fi (select all when | | |
| | | no criteria) | | |
| +------+ | | |
| | | | |
| | Timing out Flows | | |
| | Handle resource overloads | | |
+--------|---------------------------------+ +-----|-------+
| |
Flow Records (identified by Observation Domain) Flow Records
| |

Sadasivan, et al. Informational [Page 17]

RFC 5470 IPFIX Architecture March 2009

+---------+---------------------------------+
|
+--------------------|----------------------------------------------+
| | Exporting Process |
|+-------------------|-------------------------------------------+ |
|| v IPFIX Protocol | |
||+-----------------------------+ +----------------------------+| |
|||Rules for | |Functions || |
||| Picking/sending Templates | |-Packetise selected Control || |
||| Picking/sending Flow Records|->| & data Information into || |
||| Encoding Template & data | | IPFIX export packets. || |
||| Selecting Flows to export(*)| |-Handle export errors || |
||+-----------------------------+ +----------------------------+| |
|+----------------------------+----------------------------------+ |
| | |
| exported IPFIX Messages |
| | |
| +------------+-----------------+ |
| | Anonymise export packet(*) | |
| +------------+-----------------+ |
| | |
| +------------+-----------------+ |
| | Transport Protocol | |
| +------------+-----------------+ |
| | |
+-----------------------------+-------------------------------------+
|
v
IPFIX export packet to Collector

(*) indicates that the block is optional.

Figure 4: IPFIX Device functional blocks

6. Overview of the IPFIX Protocol

An IPFIX Device consists of a set of cooperating processes that
implement the functional blocks described in the previous section.
Alternatively, an IPFIX Device can be viewed simply as a network
entity that implements the IPFIX protocol. At the IPFIX Device, the
protocol functionality resides in the Exporting Process. The IPFIX
Exporting Process gets Flow Records from a Metering Process, and
sends them to the Collector(s).

At a high level, an IPFIX Device performs the following tasks:

1. Encodes Control Information into Templates.

Sadasivan, et al. Informational [Page 18]

RFC 5470 IPFIX Architecture March 2009

2. Encodes packets observed at the Observation Points into Flow
Records.

3. Packetises the selected Templates and Flow Records into IPFIX
Messages.

4. Sends IPFIX Messages to the Collector.

The IPFIX protocol communicates information from an IPFIX Exporter to
an IPFIX Collector. That information includes not only Flow Records,
but also information about the Metering Process. Such information
(referred to as Control Information) includes details of the data
fields in Flow Records. It may also include statistics from the
Metering Process, such as the number of packets lost (i.e., not
metered).

For details of the IPFIX protocol, please refer to RFC 5101 [3].

6.1. Information Model Overview

The IP Flow Information eXport (IPFIX) protocol serves for
transmitting information related to measured IP traffic over the
Internet. The protocol specification in RFC 5101 [3] defines how
Information Elements are transmitted. For Information Elements, it
specifies the encoding of a set of basic data types. However, the
list of fields that can be transmitted by the protocol, such as Flow
attributes (source IP address, number of packets, etc.) and
information about the Metering and Exporting Process (packet
Observation Point, sampling rate, Flow timeout interval, etc.), is
not specified in RFC 5101 [3]. Instead, it is defined in the IPFIX
information model in RFC 5102 [2].

The information model provides a complete description of the
properties of every IPFIX Information Element. It does this by
specifying each element's name, Field Type, data type, etc., and
providing a description of each element. Element descriptions give
the semantics of the element, i.e., say how it is derived from a Flow
or other information available within an IPFIX Device.

6.2. Flow Records

The following rules provide guidelines to be followed while encoding
the Flow Records information:

A Flow Record contains enough information so that the Collecting
Process can identify the corresponding <Per-Flow Control Information,
Configuration Control Information>.

Sadasivan, et al. Informational [Page 19]

RFC 5470 IPFIX Architecture March 2009

The Exporting Process encodes a given Information Element (as
specified in RFC 5102 [2]), based on the encoding standards
prescribed by RFC 5101 [3].

6.3. Control Information

The following rules provide guidelines to be followed while encoding
the Control Information:

o Per-Flow Control Information should be encoded such that the
Collecting Process can capture the structure and semantics of the
corresponding Flow data for each of the Flow Records exported by
the IPFIX Device.

o Configuration Control Information is conveyed to a Collector so
that its Collecting Process can capture the structure and
semantics of the corresponding configuration data. The
configuration data, which is also Control Information, should
carry additional information on the Observation Domain within
which the configuration takes effect.

For example, sampling using the same sampling algorithm, say 1 in 100
packets, is configured on two Observation Points O1 and O2. The
configuration in this case may be encoded as {ID, observation points
(O1,O2), sampling algorithm, interval (1 in 100)}, where ID is the
Observation Domain ID for the domain containing O1 and O2. The
Observation Domain ID uniquely identifies this configuration, and
must be sent within the Flow Records in order to be able to match the
right configuration control information.

The Control Information is used by the Collecting Process to:

o Decode and interpret Flow Records.

o Understand the state of the Exporting Process.

Sending Control Information from the Exporting Process in a timely
and reliable manner is critical to the proper functioning of the
IPFIX Collecting Process. The following approaches may be taken for
the export of Control Information:

1. Send all the Control Information pertaining to Flow Records prior
to sending the Flow Records themselves. This includes any
incremental changes to the definition of the Flow Records.

Sadasivan, et al. Informational [Page 20]

RFC 5470 IPFIX Architecture March 2009

2. Notify, on a near real-time basis, the state of the IPFIX Device
to the Collecting Process. This includes all changes such as a
configuration change that affects the Flow behaviour, changes to
Exporting Process resources that alter export rates, etc., which
the Collector needs to be aware of.

3. Since it is vital that a Collecting Process maintains accurate
knowledge of the Exporter's state, the export of the Control
Information should be done such that it reaches the Collector
reliably. One way to achieve this is to send the Control
Information over a reliable transport.

6.4. Reporting Responsibilities

From time to time, an IPFIX Device may not be able to observe all the
packets reaching one of its Observation Points. This could occur if
a Metering Process finds itself temporarily short of resources. For
example, it might run out of packet buffers for IPFIX export.

In such situations, the IPFIX Device should attempt to count the
number of packet losses that have occurred, and report them to its
Collector(s). If it is not possible to count losses accurately,
e.g., when transport layer (i.e., non-IPFIX) errors are detected, the
IPFIX Device should report this fact, and perhaps indicate the time
period during which some packets might not have been observed.

7. IPFIX Protocol Details

When the IPFIX Working Group was chartered, there were existing
common practices in the area of Flow export, for example, NetFlow,
CRANE (Common Reliable Accounting for Network Element), LFAP (Light-
weight Flow Admission Protocol), RTFM (Real-time Traffic Flow
Measurement), etc. IPFIX's charter required the Working Group to
consider those existing practices, and select the one that was the
closest fit to the IPFIX requirements in RFC 3917 [1]. Additions or
modifications would then be made to the selected protocol to fit it
exactly into the IPFIX architecture.

7.1. The IPFIX Basis Protocol

The Working Group went through an extensive evaluation of the various
existing protocols that were available, weighing the level of
compliance with the requirements, and selected one of the candidates
as the basis for the IPFIX protocol. For more details of the
evaluation process, please see RFC 3955 [6].

Sadasivan, et al. Informational [Page 21]

RFC 5470 IPFIX Architecture March 2009

In the basis protocol, Flow Records are defined by Templates, where a
Template is an ordered set of the Information Elements appearing in a
Flow Record, together with their field sizes within those records.

This approach provides the following advantages:

o Using the Template mechanism, new fields can be added to IPFIX
Flow Records without changing the structure of the export record
format.

o Templates that are sent to the Collecting Process carry structural
information about the exported Flow Record fields. Therefore, if
the Collector does not understand the semantics of new fields, it
can ignore them, but still interpret the Flow Record.

o Because the template mechanism is flexible, it allows the export
of only the required fields from the Flows to the Collecting
Process. This helps to reduce the exported Flow data volume and
possibly provide memory savings at the Exporting Process and
Collecting Process. Sending only the required information can
also reduce network load.

7.2. IPFIX Protocol on the Collecting Process

The Collecting Process is responsible for:

1. Receiving and decoding Flow Records from the IPFIX Devices.

2. Reporting on the loss of Flow Records sent to the Collecting
Process by an IPFIX Exporting Process.

Complete details of the IPFIX protocol are given in RFC 5101 [3].

7.3. Support for Applications

Applications that use the information collected by IPFIX may be
Billing or Intrusion Detection sub-systems, etc. These applications
may be an integral part of the Collecting Process, or they may be co-
located with the Collecting Process. The way by which these
applications interface with IPFIX systems to get the desired
information is out of scope for this document.

Sadasivan, et al. Informational [Page 22]

RFC 5470 IPFIX Architecture March 2009

8. Export Models

8.1. Export with Reliable Control Connection

As mentioned in RFC 3917 [1], an IPFIX Device must be able to
transport its Control Information and Data Stream over a congestion-
aware transport protocol.

If the network in which the IPFIX Device and Collecting Process are
located does not guarantee reliability, at least the Control
Information should be exported over a reliable transport. The Data
Stream may be exported over a reliable or unreliable transport
protocol.

Possible transport protocols include:

o SCTP: Supports reliable and unreliable transport.

o TCP: Provides reliable transport only.

o UDP: Provides unreliable transport only. Network operators would
need to avoid congestion by keeping traffic within their own
administrative domains. For example, one could use a dedicated
network (or Ethernet link) to carry IPFIX traffic from Exporter to
Collector.

8.2. Collector Failure Detection and Recovery

The transport connection (in the case of a connection-oriented
protocol) is pre-configured between the IPFIX Device and the
Collector. The IPFIX protocol does not provide any mechanism for
configuring the Exporting and Collecting Processes.

Once connected, an IPFIX Collector receives Control Information and
uses that information to interpret Flow Records. The IPFIX Device
should set a keepalive (e.g., the keepalive timeout in the case of
TCP, the HEARTBEAT interval in the case of SCTP) to a sufficiently
low value so that it can quickly detect a Collector failure. Note,
however, that extremely short keepalive intervals can incorrectly
abort the connection during transient periods of congestion. They
can also cause some level of additional network load during otherwise
idle periods.

Collector failure refers to the crash or restart of the Collecting
Process or of the Collector itself. A Collector failure is detected
at the IPFIX Device by the break in the connection-oriented transport
protocol session; depending on the transport protocol, the connection
timeout mechanisms differ. On detecting a keepalive timeout in a

Sadasivan, et al. Informational [Page 23]

RFC 5470 IPFIX Architecture March 2009

single Collector scenario, the IPFIX Device should stop sending Flow
Records to the Collector and try to reestablish the transport
connection. If Collecting Process failover is supported by the
Exporting Process, backup session(s) may be opened in advance, and
Control Information sent to the failover Collecting Process.

There could be one or more secondary Collectors with priority
assigned to them. The primary Collector crash is detected at the
IPFIX Device. On detecting loss of connectivity, the IPFIX Device
opens a Data Stream with the secondary Collector of the next highest
priority. If that secondary was not opened in advance, both the
Control Information and Data Stream must be sent to it. That
Collector might then become the primary, or the Exporting Process
might try to reestablish the original session.

8.3. Collector Redundancy

Configuring redundant Collectors is an alternative to configuring
backup Collectors. In this model, all Collectors simultaneously
receive the Control Information and Data Streams. Multiple {Control
Information, Data Stream} pairs could be sent, each to a different
Collector, from the same IPFIX Device. Since the IPFIX protocol
requires a congestion-aware transport, achieving redundancy using
multicast is not an option.

9. IPFIX Flow Collection in Special Situations

An IPFIX Device can generate, receive, and/or alter two special types
of traffic, which are listed below.

Tunnel traffic:

The IPFIX Device could be the head, midpoint, or end-point of a
tunnel. In such cases, the IPFIX Device could be handling Generic
Routing Encapsulation (GRE) [8], IPinIP [7], or Layer Two
Tunneling Protocol version 3 [9] traffic.

VPN traffic:

The IPFIX Device could be a provider-edge device that receives
traffic from customer sites belonging to different Virtual Private
Networks.

Similarly, IPFIX could be implemented on devices which perform one or
more of the following special services:

Sadasivan, et al. Informational [Page 24]

RFC 5470 IPFIX Architecture March 2009

o Explicitly drop packets. For example, a device that provides
firewall service drops packets based on some administrative
policy.

o Alter the values of fields used as IPFIX Flow Keys of interest.
For example, a device that provides NAT service can change the
source and/or destination IP address.

In cases such as those listed above, there should be clear guidelines
as to:

o How and when to classify the packets as Flows in the IPFIX Device.

o If multiple encapsulations are used to define Flows, how to convey
the same fields (e.g., IP address) in different layers.

o How to differentiate Flows based on different private domains.
For example, overlapping IP addresses in Layer-3 VPNs.

o What extra information needs to be exported so that the Collector
can make a clear interpretation of the received Flow Records.

10. Security Considerations

Flow information can be used for various purposes, such as usage-
based accounting, traffic profiling, traffic engineering, and
intrusion detection. The security requirements may differ
significantly for such applications. To be able to satisfy the
security needs of various IPFIX users, an IPFIX system must provide
different levels of security protection.

10.1. Data Security

IPFIX data comprises Control Information and Data Streams generated
by the IPFIX Device.

The IPFIX data may exist in both the IPFIX Device and the Collector.
In addition, the data is also transferred on the wire from the IPFIX
Device to the Collector when it is exported. To provide security,
the data should be protected from common network attacks.

The protection of IPFIX data within the end system (IPFIX Device and
Collector) is out of scope for this document. It is assumed that the
end system operator will provide adequate security for the IPFIX
data.

Sadasivan, et al. Informational [Page 25]

RFC 5470 IPFIX Architecture March 2009

The IPFIX architecture must allow different levels of protection to
the IPFIX data on the wire. Wherever security functions are
required, it is recommended that users should leverage lower layers
using either TLS or DTLS (Datagram Transport Layer Security), if
these can successfully satisfy the security requirement of IPFIX data
protection.

To protect the data on the wire, three levels of granularity should
be supported; these are described in the following subsections.

10.1.1. Host-Based Security

Security may not be required when the transport between the IPFIX
Device and the Collector is perceived as safe. This option allows
the protocol to run most efficiently without extra overhead, and an
IPFIX system must support it.

10.1.2. Authentication-Only

Authentication-only protection provides IPFIX users with the
assurance of data integrity and authenticity. The data exchanged
between the IPFIX Device and the Collector is protected by an
authentication signature. Any modification of the IPFIX data will be
detected by the recipient, resulting in the discarding of the
received data. However, the authentication-only option doesn't offer
data confidentiality.

The IPFIX user should not use authentication-only when sensitive or
confidential information is being exchanged. An IPFIX solution
should support this option. The authentication-only option should
provide replay attack protection. Some means to achieve this level
of security are:

o Encapsulating Security Payload (with a null encryption algorithm)

o Transport Layer Security (with a null encryption algorithm)

o IP Authentication Header

10.1.3. Encryption

Data encryption provides the best protection for IPFIX data. The
IPFIX data is encrypted at the sender, and only the intended
recipient can decrypt and have access to the data. This option must
be used when the transport between the IPFIX Device and the Collector
is unsafe, and the IPFIX data needs to be protected. It is

Sadasivan, et al. Informational [Page 26]

RFC 5470 IPFIX Architecture March 2009

recommended that the underlying transport layer's security functions
be used for this purpose. Some means to achieve this level of
security are:

o Encapsulating Security Payload

o Transport Layer Security Protocol

The data encryption option adds overhead to the IPFIX data transfer.
It may limit the rate that an Exporter can report its Flow Records to
the Collector, due to the resource requirement for running
encryption.

10.2. IPFIX End-Point Authentication

It is important to make sure that the IPFIX Device is talking to the
"right" Collector rather than to a masquerading Collector. The same
logic also holds true from the Collector's point of view, i.e., it
may want to make sure it is collecting the Flow Records from the
"right" IPFIX Device. An IPFIX system should allow the end-point
authentication capability so that either one-way or mutual
authentication can be performed between the IPFIX Device and
Collector.

The IPFIX architecture should use any existing transport protection
protocols, such as TLS, to fulfil the authentication requirement.

10.3. IPFIX Overload

An IPFIX Device could become overloaded under various conditions.
This may be because of exhaustion of internal resources used for Flow
generation and/or export. Such overloading may cause loss of data
from the Exporting Process, either from lack of export bandwidth
(possibly caused by an unusually high number of observed Flows) or
from network congestion in the path from Exporter to Collector.

IPFIX Collectors should be able to detect the loss of exported Flow
Records, and should at least record the number of lost Flow Records.

10.3.1. Denial-of-Service (DoS) Attack Prevention

Since one of the potential usages for IPFIX is for intrusion
detection, it is important for the IPFIX architecture to support some
kind of DoS resistance.

Sadasivan, et al. Informational [Page 27]

RFC 5470 IPFIX Architecture March 2009

10.3.1.1. Network under Attack

The network itself may be under attack, resulting in an overwhelming
number of IPFIX Messages. An IPFIX system should try to capture as
much information as possible. However, when a large number of IPFIX
Messages are generated in a short period of time, the IPFIX system
may become overloaded.

10.3.1.2. Generic DoS Attack on the IPFIX Device and Collector

The IPFIX Device and Collector may be subject to generic DoS attacks,
just as any system on any open network. These types of attacks are
not IPFIX specific. Preventing and responding to such types of
attacks are out of the scope of this document.

10.3.1.3. IPFIX-Specific DoS Attack

There are some specific attacks on the IPFIX portion of the IPFIX
Device or Collector:

o The attacker could overwhelm the Collector with spoofed IPFIX
Export packets. One way to solve this problem is to periodically
synchronise the sequence numbers of the Flow Records between the
Exporting and Collecting Processes.

o The attacker could provide false reports to the Collector by
sending spoofed packets.

The problems mentioned above can be solved to a large extent if the
control packets are encrypted both ways, thereby providing more
information that the Collector could use to identify and ignore
spoofed data packets.

11. IANA Considerations

The IPFIX Architecture, as set out in this document, has two sets of
assigned numbers, as outlined in the following subsections.

11.1. Numbers Used in the Protocol

IPFIX Messages, as described in RFC 5101 [3], use two fields with
assigned values. These are the IPFIX Version Number, indicating
which version of the IPFIX Protocol was used to export an IPFIX
Message, and the IPFIX Set ID, indicating the type for each set of
information within an IPFIX Message.

Sadasivan, et al. Informational [Page 28]

RFC 5470 IPFIX Architecture March 2009

Values for the IPFIX Version Number and the IPFIX Set ID, together
with the considerations for assigning them, are defined in RFC 5101
[3].

11.2. Numbers Used in the Information Model

Fields of the IPFIX protocol carry information about traffic
measurement. They are modelled as elements of the IPFIX information
model RFC 5102 [2]. Each Information Element describes a field that
may appear in an IPFIX Message. Within an IPFIX Message, the field
type is indicated by its Field Type.

Values for the IPFIX Information Element IDs, together with the
considerations for assigning them, are defined in RFC 5102 [2].

12. Acknowledgements

The document editors wish to thank all the people contributing to the
discussion of this document on the mailing list, and the design teams
for many valuable comments. In particular, the following made
significant contributions:

Tanja Zseby
Paul Calato
Dave Plonka
Jeffrey Meyer
K.C.Norseth
Vamsi Valluri
Cliff Wang
Ram Gopal
Jc Martin
Carter Bullard
Reinaldo Penno
Simon Leinen
Kevin Zhang
Paul Aitken
Brian Trammell

Special thanks to Dave Plonka for the multiple thorough reviews.

Sadasivan, et al. Informational [Page 29]

RFC 5470 IPFIX Architecture March 2009

13. References

13.1. Normative References

[1] Quittek, J., Zseby, T., Claise, B., and S. Zander, "Requirements
for IP Flow Information Export (IPFIX)", RFC 3917, October 2004.

[2] Quittek, J., Bryant, S., Claise, B., Aitken, P., and J. Meyer,
"Information for Model IP Flow Information Export", RFC 5102,
January 2008.

[3] Claise, B., "Specification of the IP Flow Information Export
(IPFIX) Protocol for the Exchange of IP Traffic Flow
Information", RFC 5101, January 2008.

[4] Zseby, T., Boschi, E., Brownlee, N., and B. Claise, "IPFIX
Applicability", RFC 5472, March 2009.

13.2. Informative References

[5] Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson,
"RTP: A Transport Protocol for Real-Time Applications", STD 64,
RFC 3550, July 2003.

[6] Leinen, S., "Evaluation of Candidate Protocols for IP Flow
Information Export (IPFIX)", RFC 3955, October 2004.

[7] Simpson, W., "IP in IP Tunneling", RFC 1853, October 1995.

[8] Farinacci, D., Li, T., Hanks, S., Meyer, D., and P. Traina,
"Generic Routing Encapsulation (GRE)", RFC 2784, March 2000.

[9] Lau, J., Townsley, M., and I. Goyret, "Layer Two Tunneling
Protocol - Version 3 (L2TPv3)", RFC 3931, March 2005.

Sadasivan, et al. Informational [Page 30]

RFC 5470 IPFIX Architecture March 2009

Authors' Addresses

Ganesh Sadasivan
Rohati Systems
1192 Borregas Ave.
Sunnyvale, CA 94089
USA

EMail: [email protected]

Nevil Brownlee
CAIDA | The University of Auckland
Private Bag 92019
Auckland 1142
New Zealand

Phone: +64 9 373 7599 x88941
EMail: [email protected]

Benoit Claise
Cisco Systems, Inc.
De Kleetlaan 6a b1
1831 Diegem
Belgium

Phone: +32 2 704 5622
EMail: [email protected]

Juergen Quittek
NEC Laboratories Europe, NEC Europe Ltd.
Kurfuersten-Anlage 36
Heidelberg 69115
Germany

Phone: +49 6221 4342-115
EMail: [email protected]
URI: http://www.neclab.eu/

Sadasivan, et al. Informational [Page 31]