Internet DRAFT - draft-akhter-opsawg-perfmon-method
draft-akhter-opsawg-perfmon-method
Network Working Group A. Akhter
Internet-Draft Cisco Systems
Intended status: Standards Track March 27, 2012
Expires: September 28, 2012
Methodology for Network Flow Performance Measurement
draft-akhter-opsawg-perfmon-method-02.txt
Abstract
There is a need to be able to quantify and report the performance of
network applications and the network service in handling user data.
This performance data provides information essential in validating
service level agreements, fault isolation as well as early warnings
of network greater problems. This document describes a generic
methodology for calculating metrics related to network based
applications. In addition, to the performance metrics, several
additional information elements are included to help provide greater
context to the reports. The measurements use audio/video
applications as base examples but are not restricted to these class
of applications.
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on September 28, 2012.
Copyright Notice
Copyright (c) 2012 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
(http://trustee.ietf.org/license-info) in effect on the date of
Akhter Expires September 28, 2012 [Page 1]
Internet-Draft PerfMon Meth March 2012
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. General Usage . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1. Quality of Service (QoS) Monitoring . . . . . . . . . . . 5
3.2. Service Level Agreemnt (SLA) Validation . . . . . . . . . 5
3.3. Fault Isolation and Troubleshooting . . . . . . . . . . . 5
4. New Information Elements . . . . . . . . . . . . . . . . . . . 6
4.1. Transport Layer . . . . . . . . . . . . . . . . . . . . . 6
4.1.1. perfPacketLoss . . . . . . . . . . . . . . . . . . . . 6
4.1.2. perfPacketExpected . . . . . . . . . . . . . . . . . . 8
4.1.3. perfPacketLossRate . . . . . . . . . . . . . . . . . . 9
4.1.4. perfPacketLossEvent . . . . . . . . . . . . . . . . . 10
4.1.5. perfPacketInterArrivalJitterAvg . . . . . . . . . . . 11
4.1.6. perfPacketInterArrivalJitterMin . . . . . . . . . . . 12
4.1.7. perfPacketInterArrivalJitterMax . . . . . . . . . . . 13
4.1.8. perfRoundTripNetworkDelay . . . . . . . . . . . . . . 13
4.2. User and Application Layer . . . . . . . . . . . . . . . . 14
4.2.1. perfSessionSetupDelay . . . . . . . . . . . . . . . . 14
4.3. Contextual Elements . . . . . . . . . . . . . . . . . . . 15
4.3.1. mediaRTPSSRC . . . . . . . . . . . . . . . . . . . . . 15
4.3.2. mediaRTPPayloadType . . . . . . . . . . . . . . . . . 16
4.3.3. mimeType . . . . . . . . . . . . . . . . . . . . . . . 16
5. Security Considerations . . . . . . . . . . . . . . . . . . . 17
6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 17
7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.1. Normative References . . . . . . . . . . . . . . . . . . . 18
7.2. Informative References . . . . . . . . . . . . . . . . . . 18
Author's Address . . . . . . . . . . . . . . . . . . . . . . . . . 20
Akhter Expires September 28, 2012 [Page 2]
Internet-Draft PerfMon Meth March 2012
1. Introduction
Today's networks support a multitude of highly demanding and
sensitive network applications. Network issues are readily apparent
by the users of these applications due to the sensitivity of these
applications to impaired network conditions. Examples of these
network applications include applications making use of IP based
audio, video, database transactions, virtual desktop interface (VDI),
online gaming, cloud services and many more. In some cases, the
impaired application translates directly to loss of revenue. In
other cases, there may be regulatory or contractual service level
agreements that motivate the network operator. Due to the
sensitivity of these types of applications to impaired service, it
leaves a poor impression of the network service on the user--
regardless of the actual performance of the network itself. In the
case of an actual problem within the network service, monitoring the
performance may yield an early indicator of a much more serious
problem.
Due to the demanding and sensitive nature of these applications,
network operators have tried to engineer their networks towards
wringing better and differentiated performance. However, that same
differentiated design prevents network operators from extrapolating
observational data from one application to another, or from one set
of synthetic (active test) test traffic to actual application
performance. This gap highlights the importance of generic
measurements as well as the reliance on user traffic measurents--
rather than synthetic tests.
Performance measurements on user data provide greater visibility not
only into the quality of experience of the end users but also
visibility into network health. With regards to network health, as
flow performance is being measured, there will be visibility into the
end to end performance which means that not only visibility into
local network health, but also viability into remote network health.
If these measurements are made at multiple points within the network
(or between the network and end device) then there is not only
identification that there might be an issue, but a span of area can
be established where the issue might be. The resolution of the fault
increases with the number of measurement points along the flow path.
The IP Flow Information Export Protocol (IPFIX) [RFC5101] provides
new levels of flexibility in reporting from measurement points across
the life cycle of a network based application. IPFIX can provide
granular results in terms of flow specificity as well as time
granularity. At the same time, IPFIX allows for summarization of
data along different types of boundaries for operators that are
unconcerned about specific sessions but about health of a service or
Akhter Expires September 28, 2012 [Page 3]
Internet-Draft PerfMon Meth March 2012
a portion of the network. This document details the methodlogy of
measurement, while an accompanying document describes the expression
of the measurements in IPFIX format.
Where possible, an attempt has been made to make use of existing
definitions of metrics ([RFC4710]) and if needed, clarify and expand
on them to widen their usage with additional applications, and
network devices. For example, the RTP measurments have generally
defined from the prespective of end systems rather than intermdiate
nodes which are not alwyas privy to the application context and may
have limited scaling properties. The methodology described in
[I-D.ietf-pmol-sip-perf-metrics] is used to describe the methodology
of measurement.
There has been related work in this area such as [RFC2321].
[I-D.huici-ipfix-sipfix], and [VoIP-monitor]. This document is also
an attempt to generalize as well as standardize the reporting formats
and measurement methodology.
2. Terminology
Terms used in this document that are defined in the Terminology
section of the IPFIX Protocol [RFC5101] document are to be
interpreted as defined there.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [RFC2119].
In addition, the information element definitions use the following
terms:
Name: Name of the information element
Description: Short description of what the information element is
trying to convey.
Observation Point: Where the measurement is meant to be performed.
Either at an intermediate system (for example, a router) or end
system.
Use and Applications An explanation of how this particular
information element would be used.
Akhter Expires September 28, 2012 [Page 4]
Internet-Draft PerfMon Meth March 2012
Calculation Method: In the case of metrics, this section describes
how the metric is calculated, as well as any special conditions.
Units of Measurement: In the case of metrics, what are the units of
measurement. The text here is expected to be wider and more
descriptive than in the IPFIX Element Units section.
Measurement Timing: Discussion on the acceptable range of timing and
sampling intervals.
3. General Usage
3.1. Quality of Service (QoS) Monitoring
The network operator needs to be able to gauge the end user's
satisfaction with the network service. While there are many
components of the satisfaction such as pricing, packaging, offering,
etc., a major component of satisfaction is delivering a consistent
service. The user builds trust on this consistency of the network
service and runs network applications with confidence-- which is of
course the end goal. Without the ability to deliver a consistent
service for end user network applications network operator will be
left dealing with price sensitive disgruntled users with very low
expectations and utilization (if they don't have choice of operator)
or abandonment (if they have choice).
3.2. Service Level Agreemnt (SLA) Validation
Similar to QoS and QoE validation, there might be contractual or
regulatory requirements that need to be met by the network operator.
Monitoring the performance of the flows allows the application
operator, network operator as well as the end user to validate if the
target service is being delivered. While there is quite a diversity
in the codification of network SLAs, the eventually involve some
measurement of network uptime, end to end latency, end to end jitter
and perhaps service response time. In the case of SLA violation, the
start and end times, nature and network scope of the violation needs
to be captured to allow for the most accurate settling of the SLA
violation.
3.3. Fault Isolation and Troubleshooting
It has been generally easier to troubleshoot and fix problems that
are binary in nature: it either works or does not work. The host is
pingable or not pingable. However, it is the much more difficult to
resolve transitory issues that move from location to location, are
not complete faiures and sometimes with unverfifiable end user
Akhter Expires September 28, 2012 [Page 5]
Internet-Draft PerfMon Meth March 2012
reports as the only indication of a problem. It is these
intermittent and seemingly inconsistent network impairments that
performance metrics can be extremely helpful with. Just the basic
timely detection that there is a problem (or an impending problem)
can give the operator provider the confidence that there is a real
problem that needs to be resolved. The next step would be to assist
the operator in a speedy resolution by providing information
regarding the network location and nature of the problem.
4. New Information Elements
The information elements are organized into two main groups:
Transport Layer: Metrics that might be calculated from observations
at higher layers but essentially provide information about the
network transport of user date. For example, the metrics related
to packet loss, latency and jitter would be defined here.
User and Application Layer: Metrics that are might be affected by
the network indirectly, but are ultimately related to user, end-
system and session states. For example, session setup time,
transaction rate and session duration would be defined here.
Contextual Elements: Information elements that provide further
context to the metrics. For example, media type, codec type, and
type of application would be defined here.
4.1. Transport Layer
4.1.1. perfPacketLoss
Name: perfPacketLoss
Description: The packet loss metric reports the number of individual
packets that were lost in the reporting interval.
Observation Point: The observation can be made anywhere along the
media path or on the endpoints them selves. The observation is
only relevant in a unidirectional sense.
Use and Applications The packet loss metric can be used to determine
if there is a network impairment that is causing packet loss
upstream of the measurement point. When there are observation
points on either side of the impairment location it is possible to
locate the impairment. With the location information the operator
can is able to perform quicker fault-isolation as well as shorten
time to resolution. Depending on implementation and operator
Akhter Expires September 28, 2012 [Page 6]
Internet-Draft PerfMon Meth March 2012
configuration, the granularity of contextual information can be
very specific. For example, these traffic loss statistics when
sent with IP subnet or DSCP information can provide visibilty into
QoS specific or network topology issues.
Calculation Method: This metric requires that each IP packet be
individually marked with a monotonically incrementing sequence
number. A number of encapsulations support this type of
sequencing: IPSec ESP [RFC4303], GRE [RFC2890] and RTP [RFC3550].
An analysis of the sequence number field can yield the lost number
of packets. In certain cases, there might be an element of
discovery and synchronization of the flow itself before the
measurement can be made. An example of this can be found for RTP
flows running on ephemeral UDP port numbers. In these cases,
reporting 0 as packet loss would be misleading and the value
0xFFFFFFFF MUST be used in cases where the packet loss value
cannot be determined. In the case of a monitor interval where
synchronization was achieved mid-interval, the loss packet counter
MAY be used to represent the remainder of the interval. As this
metric is a deltaCounter, the number of loss packets only
represent the observation within the reporting interval. Due to
the dependency on the arrival of a packet with a sequence number
to calculate loss, the loss calculation may be indefinitely
delayed if no more packets arrive at all. For the case of RTP, in
addition to the 16 bit sequence number field in RFC3550, there is
also the additional 16-bit high-order sequence number field (for a
total of 32-bit seq number space) that is used in RFC3497
[RFC3497]. RFC3497 traffic runs at a very high rate and the 32-
bit field allow for additional time for wrapping (21 seconds).
So, a loss span of greater than 21 seconds measured only by the
16-bit field will lead to inaccurate reporting. In the case of
secure RTP [RFC3711], the relevant portion of the RTP header is in
the clear and lost packet counting can still be performed. It is
important to note that the sequence number space is unique per RTP
SSRC. Therefore it is important to track the high sequence number
seen on a per SSRC-5-tuple basis. There may be multiple SSRCs in
a single 5-tuple. Certain applications inject non-RTP traffic
into the same 5-tuple as the media stream. RTCP packets may be
seen in the same 5-tuple as the RTP stream [RFC5761], and STUN
[RFC5389] packets may also be seen. The loss detection should
ignore these packets. There may be spans within the network where
header compression schemes such as [RFC2508] are used. In cases
where the measurement device is terminating the compression, and
the measurement implementation does not support calculation of the
metric the value 0xFFFFFFFF MUST be reported. In other cases the
measurement point may be at a midpoint of the header compression
network span. Depending on the mechanics of header compression,
sequencing information may be present and it is possible to
Akhter Expires September 28, 2012 [Page 7]
Internet-Draft PerfMon Meth March 2012
calculate the metric. In such cases the implementation SHOULD
perform the calculation and report the metric.
Units of Measurement: packets
Measurment Timing To be able to calculate this metric a continuous
set of the flow's packets (as each would have an incrementing
sequence number) needs to be monitored. Therefore, per-packet
sampling would prevent this metric from being calculated.
However, there are other sampling methodologies that might be
usable. It is possible to generate sampled metrics by sampling
spans of continuous packets, however a portion of the span may
have to be utilized for resynchronization of the sequence number.
Another form of acceptable sampling would be at the flow level.
4.1.2. perfPacketExpected
Name: perfPacketExpected
Description: The number of packets there were expected within a
monitoring interval.
Observation Point: The observation can be made anywhere along the
media path or on the endpoints them selves. The observation is
only relevant in a unidirectional sense.
Use and Applications The perfPacketExpected is a mid-calculation
metric used in the generation of perfPacketLossRate. It is
equivilent to the highest received packet sequence number at the
time of measurement. As the value only increments when packets
are received, packet loss may be occouring at the time of
measurement but perfPacketExpected remains constant.
Calculation Method: The subtraction of the last sequence number from
the first sequence number in monitoring interval yields the
expected count. As discussed with perfPacketLost, there might be
a delay due to synchronization with the flow's sequence numbers
and in such times the value of the metric should be set to
0xFFFFFFFE. Care has to be taken to account for cases where the
packet's sequence number field wraps. For RTP, the expected count
calculation formula can be found in Appendix A.3 of [RFC3550].
Refer to the perfPacketLoss metric regarding considerations for
header compression. The value 0xFFFF is used to represent cases
where the metric could not be calculated.
Akhter Expires September 28, 2012 [Page 8]
Internet-Draft PerfMon Meth March 2012
Units of Measurement: packets
Measurment Timing Same considerations as perfPacketLoss
4.1.3. perfPacketLossRate
Name: perfPacketLossPercentage
Description: Percentage of number of packets lost out of the total
set of packets sent.
Observation Point: The observation can be made anywhere along the
media path or on the endpoints them selves. The observation is
only relevant in a unidirectional sense.
Use and Applications The perfPacketLossRate metric can be used to
normalize the perfPacketLoss metric to handle cases where
different flows are running at different packet per second (PPS)
rates. Due to the normalization, comparisons can now be made
against thresholds (for creating alerts, etc.). In addition, the
percentage form of the metric allows for comparisons against other
flows at the same observation point to determine if there is an
equal bias for drops between the flows. Otherwise, the
perfoPacketLossRate is used in same way as perfPacketLoss. This
value can be derived from perfPacketExpected and perfPacketLoss
and is offered as a convenience to ease functions such as
thresholding, and pre-computed reporting. It shoudl be noted that
for large values of perfPacketExpected and perfPacketLoss it might
be preferable and more accurate for the conversion to percentage
to occour at a later stage where the accuracy can be controlled.
Calculation Method: The number of lost packets divided by the number
of expected packets in an interval period multiplied by 100. In
cases where perfPacketLoss is unknown (for example due to
synchronization issues), the perfPacketLossRate would also be
unknown. If there are multiple flows whose loss rate is being
aggregated, then the average of the individual flows is used.
Refer to the perfPacketLoss metric regarding considerations for
header compression.
Units of Measurement: percentage
Measurment Timing Same notes as perfPacketLossPercentage
Akhter Expires September 28, 2012 [Page 9]
Internet-Draft PerfMon Meth March 2012
4.1.4. perfPacketLossEvent
Name: perfPacketLossEvent
Description: The packet loss event metric reports the number of
continuous sets of packets that were lost in the reporting
interval.
Observation Point: The observation can be made anywhere along the
media path or on the endpoints them selves. The observation is
only relevant in a unidirectional sense.
Use and Applications The perfPacketLossEvent metric can provide loss
information for protocols that do not implement per packet
sequencing. Similarly to the perfPacketLoss metric, the packet
loss event metric can be used to determine if there is a network
impairment that is causing packet loss upstream of the measurement
point. In cases where both the perfPacketLoss and
perfPacketLossEvent metric are available, the ratio between the
packet loss and packet event count can provide the average loss
length. The average loss length provides additional information
regarding the cause of the loss. For example, a dirty fiber
connection might have a low average loss length, while a routing
protocol convergence will have a high loss length.
Calculation Method: This data value is a simplified version of the
Lost Packets metric. Whereas Lost Packets counts individual
packet loss, the 'loss event count' metric counts sets of packets
that are lost. For example, in the case of a sequence of packets:
1,3,6,7,10 the packets marked 2,4,5,8 and 9 are lost. So, a total
of 5 packets are lost. This same sequence translates to 3 loss
events: (2), (4,5) and (8,9). In the case of RTP, the sequence
number in the RTP header can be used to identify loss events.
Certain protocols such as TCP and UDP+MPEG2-TS encapsulation in IP
have sequencing information, but the sequence field is incremented
by individual IP packets. As a side note, in the case of UDP+
MPEG2-TS encapsulation the simple use of RTP+MPEG2-TS via
[RFC2250] results in the avaliability of the more granular
perfPacketLoss metrics. In these cases, the perfPacketLoss metric
cannot be calculated but the perfPacketLossEvent can be calculated
and can provide detection of loss. The value 0xFFFFFFFF is used
to represent non-applicable cases such as lack of sequence number
synchronization. Many of the same considerations as for
perfPacketLoss apply to perfPacketLoss event. Please refer to the
Calculation Method section of the perfPacketLoss.
Akhter Expires September 28, 2012 [Page 10]
Internet-Draft PerfMon Meth March 2012
Units of Measurement: event counts
Measurment Timing Please refer to the measurement timing section of
perfPacketLoss.
4.1.5. perfPacketInterArrivalJitterAvg
Name: perfPacketInterArrivalJitterAvg
Description: This metric measures the absolute deviation of the
difference in packet spacing at the measurement point compared to
the packet spacing at the sender.
Observation Point: The observation can be made anywhere along the
media path or on the receiver. The observation is only relevant
in a unidirectional sense.
Use and Applications The inter arrival jitter data value can be used
be network operator to determine the network's impact to the
spacing in between a media stream's packets as they traverse the
network. For example, in the case of media applications, the
receiving end system is expecting these packets to come in at a
particular periodicity and large deviations may result in de-
jitter buffers adding excessive delay, or the media packets being
discarded. When the data is reported from multiple intermediate
nodes, the area of the network that is having a detrimental
contribution can be identified. On a non-media application level,
the inter arrival jitter metrics can be used for early indication
queuing contention within the network (which could lead to packet
loss).
Calculation Method: The inter arrival jitter value makes use of the
association of sending time with an IP packets and comparison of
the arrival time on the monitoring point. In certain protocols, a
representation of sending time is encoded into the header itself.
For example, in the case of RTP packets, the RTP header's
timestamps field represents encoder clock ticks-- which are
representations of time. Similarly, in the case of TCP options
encode absolute timestamps values. For RTP the calculation method
can be found in Appendix A of [RFC3550]. It should be noted that
the RFC3550 calculation is on the last 16 packets measured. The
most recent value calculated SHOULD be reported at the end of the
monitoring interval. The range of the jitter values during the
monitoring interval can be reported using
perfPacketInterArrivalJitterMin and
perfPacketInterArrivalJitterMax. Similarly to the perfPacketLoss
case there may be periods of time where the jitter value cannot be
calculated. In these cases, the 0xFFFFFFFF value should be used
Akhter Expires September 28, 2012 [Page 11]
Internet-Draft PerfMon Meth March 2012
to convey the lack of availability of the metric. As mentioned
earlier, the RTP header timestamps is actually a 'sample-stamp'
(ie clicks) from the encoder's clock. The frequency of the clock
is dependent on the codec. Some codecs (eg AAC-LD) support
multiple possible frequencies one of which is then selected for
the media-stream. The mapping to clock rate can be performed via
mapping from the static RTP payload type (RTP-PT), but newer
codecs are make use of the dynamic payload type range and the
RTP-PT (in the dynamic case) cannot be used to determine the clock
frequency. There are various methods by which the clock frequency
(deep packet inspection of the signalling, manual configuration,
etc.) can be associated to the calculation method. The frequency
should be locked in the metering layer to a unique combination of
the IP source, IP destination, IP protocol layer-4 ports, RTP-PT
and SSRC. By strict RFC3550 definition, the SSRC is set to a
specific encoder clock and it is the SSRC that should be tracked
rather than payload type. However, in recent discussions it has
been noted that there are RTP implementations that might change
the encoder clock frequency while maintaining the SSRC value. An
encoder frequency change will be accompanied by a different
RTP-PT.
Units of Measurement: microseconds
Measurment Timing Please refer to the measurement timing section of
perfPacketLoss.
4.1.6. perfPacketInterArrivalJitterMin
Name: perfPacketInterArrivalJitterMin
Description: This metric measures the minimum value the calculation
used for perfPacketInterArrivalJitterAvg within the monitoring
interval.
Observation Point: The observation can be made anywhere along the
media path or on the receiver. The observation is only relevant
in a unidirectional sense.
Use and Applications Please refer to the 'Use and Applications'
section of perfPacketInterArrivalJitterAvg. This specific metric,
along with perfPacketInterArrivalJitterMax, is to capture the
range of measurements observed within a monitoring interval as the
average function may hide extremes.
Akhter Expires September 28, 2012 [Page 12]
Internet-Draft PerfMon Meth March 2012
Calculation Method: Please see the perfPacketInterArrivalJitterAvg
section for general calculation section. The average calculation
is evaluated on a running basis over the last 16 packets and the
entire monitoring interval is not covered. In this metric, the
minimum value is taken over the entire monitoring interval.
Units of Measurement: microseconds
Measurment Timing Please refer to the measurement timing section of
perfPacketLoss.
4.1.7. perfPacketInterArrivalJitterMax
Name: perfPacketInterArrivalJitterMax
Description: This metric measures the maximum value the calculation
used for perfPacketInterArrivalJitterAvg within the monitoring
interval.
Observation Point: The observation can be made anywhere along the
media path or on the receiver. The observation is only relevant
in a unidirectional sense.
Use and Applications Please refer to the 'Use and Applications'
section of perfPacketInterArrivalJitterAvg. This specific metric,
along with perfPacketInterArrivalJitterMin, is to capture the
range of measurements observed within a monitoring interval as the
average function may hide extremes.
Calculation Method: Please see the perfPacketInterArrivalJitterAvg
section for general calculation section. The average calculation
is evaluated on a running basis over the last 16 packets and the
entire monitoring interval is not covered. In this metric, the
maximum value is taken over the entire monitoring interval.
Units of Measurement: microseconds
Measurment Timing Please refer to the measurement timing section of
perfPacketLoss.
4.1.8. perfRoundTripNetworkDelay
Name: perfRoundTripNetworkDelay
Description: This metric measures the network round trip time
between end stations for a flow.
Akhter Expires September 28, 2012 [Page 13]
Internet-Draft PerfMon Meth March 2012
Observation Point: The observation can be made anywhere along the
flow path as long as the bidirectional network delay is accounted
for.
Use and Applications The perfRoundTripNetworkDelay metric can be
used in multiple ways. If the applicaiton being monitored
provides interactive feedback to the user the
perfRoundTripNetworkDelay can be used to judge the 'liveliness' of
the application experience. Other use cases may involve
troubleshooting throughput issues where the transport protocol's
throughtput is affected by network delay.
Calculation Method: perfRoundTripNetworkDelay can estimated by
accounting for the network flight time between a transport
protocol request and response. In the case of TCP, this would the
time difference between the TCP SYN and ACK packets in the TCP
handshake. It should be noted that at times other than the TCP
handshake the time difference between TCP end station packet. For
RTP (RFC3550) based applications, the network round trip can be
calculated by analysis of hte RTCP sending and receive times.
Units of Measurement: microseconds
Measurment Timing Depending on the method used to calculate the
round trip time, the measurment may only be possible at specific
times during the session lifecycle. In time periods where the
metric is not current 'not calculated' SHOULD be reported.
4.2. User and Application Layer
4.2.1. perfSessionSetupDelay
Name: perfSessionSetupDelay
Description: The Session Setup Delay metric reports the time taken
from a request being initiated by a host/endpoint to the response
(or request indicator) to the request being observed. This metric
is defined in [RFC4710], however the units have been updated to
microseconds.
Observation Point: This metric needs to be calculated where both
request and response can be observed. This could be at network
choke points, application proxies, or within the end systems
themselves.
Akhter Expires September 28, 2012 [Page 14]
Internet-Draft PerfMon Meth March 2012
Use and Applications The session setup delay metric can measure the
end user initial wait experience as seen from the network
transaction level. The value will not only include the network
flight time, but also includes the server response time and may be
used to alert the operator in cases where the overall service is
overloaded and thus sluggish, or within normal operating values.
Calculation Method: Measure distance in time between the first bit
of request and the first bit of the response. For the case of
SIP, please see Section 4.3.1 of [I-D.ietf-pmol-sip-perf-metrics]
Units of Measurement: microseconds
Measurment Timing This measurement can be sampled on a session by
session basis. It may be advisable to set sample targets on a per
source range - to destination basis. Due to the nature of
measurement intervals, there may be a period of time (and thus
measurement reports) in which the perfSessionSetupDelay value has
not been calculated. In these cases the value 0xFFFFFFFE MUST be
used and can be interpreted to mean not applicable. For
measurement intervals after perfSessionSetupDelay has been
calculated and the existing calculated perfSessionSetupDelay value
SHOULD be sent if reporting only on that single session. However,
if multiple sessions are summarized in the report then the average
for perfSessionSetupDelay values calculated in the most recent
interval SHOULD be used. The intention with this behavior is to
acknowledge that the value has not bee calculated, and when it has
provide the freshest values available.
4.3. Contextual Elements
4.3.1. mediaRTPSSRC
Name: mediaRTPSSRC
Description: Value of the synchronization source (SSRC) field in the
RTP header of the flow. This field is defined in [RFC3550]
Observation Point: This metric can be gleaned from the RTP packets
directly, so the observation point needs to on the flow path or
within the endpoints.
Use and Applications The RTP SSRC value denotes a specific media
stream. As such when trying to differentiate media stream
problems between session participants the SSRC field is needed.
Akhter Expires September 28, 2012 [Page 15]
Internet-Draft PerfMon Meth March 2012
Calculation Method: Copy from RTP header's SSRC field as defined in
[RFC3550]. In the case of a non-RTP flow, or the time period in
which the flow has not been verified to be a RTP flow the value
0xFFFFFFFE MUST be reported.
Units of Measurement: identifier
Measurment Timing It is possible that the SSRC may have be
renegotiated mid-session due to collisions with other RTP senders.
4.3.2. mediaRTPPayloadType
Name: mediaRTPPayloadType
Description: The value of the RTP Payload Type Field as seen in the
RTP header of the flow. This field is defined in [RFC3550]
Observation Point: This metric can be gleaned from the RTP packets
directly, so the observation point needs to on the flow path or
within the endpoints.
Use and Applications The RTP PT conveys the payload format and media
encoding used in the RTP payload. For simple cases, where the RTP
PT is from the statically defined range this can lead to an
understanding of type of media codec used. With the knowledge of
the codec being used the degree of media impairment (given loss
values and jitter) can be estimated better. However, for more
recent codecs, the RTP dynamic range is used. In these cases the
RTP payload values are dynamically negotiated. In the case of a
non-RTP flow, or the time period in which the flow has not been
verified to be a RTP flow, the value 0xFFFF MUST be reported.
Calculation Method: Copy from RTP header's RTP-PT field as defined
in [RFC3550]
Units of Measurement: identifier
Measurment Timing
4.3.3. mimeType
Name: mimeType
Description: The mime type describes the content of the flow.
Akhter Expires September 28, 2012 [Page 16]
Internet-Draft PerfMon Meth March 2012
Observation Point: The ideal location of this metric is on the
application generators and consumers. However, given application
signalling inspection or static configuration it is possible that
intermediate nodes are able to generate mime type (eg. codec name)
information.
Use and Applications The mime type value conveys information
regarding the content of a flow. For example, in the case of
Audio/Video applications the name of the codec used to encode the
media in the flow. Simply reporting loss and jitter measurements
are useful for detection of network problems. However, judging
the degree of the impact on the audio/video experience needs
additional information. The most basic information is the codec
being used which when coupled with per-codec knowledge of
sensitivity to the transport metrics a better idea of the
experience can be gained.
Calculation Method: The valid values for the mime type are listed on
the IANA mime type registry. For Audio/Video codecs, there is a
specific media-types registry. Analysis of the RTP payload type
may lead to the determination of the media codec. However, with
the use of the RTP dynamic payload type range the media
information is not encoded into the data packet. For these cases,
intermediate nodes may need to perform inspection of the
signalling (SIP, H.323, RTSP, etc.). In cases where the
mediaCodec cannot be determined, the value 'unknown' MUST be used.
Units of Measurement: identifier
Measurment Timing
5. Security Considerations
The recommendations in this document do not introduce any additional
security issues to those already mentioned in [RFC5101] and [RFC5477]
6. Acknowledgements
The authors would like to thank Rahul Patel, Jan Novak, Al Morton,
Brad Fawcett, Doug Manley and Shingo Kashima for their invaluable
review and comments. Thank-you.
7. References
Akhter Expires September 28, 2012 [Page 17]
Internet-Draft PerfMon Meth March 2012
7.1. Normative References
[RFC5101] Claise, B., "Specification of the IP Flow Information
Export (IPFIX) Protocol for the Exchange of IP Traffic
Flow Information", RFC 5101, January 2008.
[RFC5610] Boschi, E., Trammell, B., Mark, L., and T. Zseby,
"Exporting Type Information for IP Flow Information Export
(IPFIX) Information Elements", RFC 5610, July 2009.
[RFC4710] Siddiqui, A., Romascanu, D., and E. Golovinsky, "Real-time
Application Quality-of-Service Monitoring (RAQMON)
Framework", RFC 4710, October 2006.
[RFC5102] Quittek, J., Bryant, S., Claise, B., Aitken, P., and J.
Meyer, "Information Model for IP Flow Information Export",
RFC 5102, January 2008.
[RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V.
Jacobson, "RTP: A Transport Protocol for Real-Time
Applications", STD 64, RFC 3550, July 2003.
[RFC3497] Gharai, L., Perkins, C., Goncher, G., and A. Mankin, "RTP
Payload Format for Society of Motion Picture and
Television Engineers (SMPTE) 292M Video", RFC 3497,
March 2003.
[RFC5389] Rosenberg, J., Mahy, R., Matthews, P., and D. Wing,
"Session Traversal Utilities for NAT (STUN)", RFC 5389,
October 2008.
[I-D.ietf-pmol-sip-perf-metrics]
Malas, D. and A. Morton, "Basic Telephony SIP End-to-End
Performance Metrics", draft-ietf-pmol-sip-perf-metrics-07
(work in progress), September 2010.
[iana-ipfix-assignments]
Internet Assigned Numbers Authority, "IP Flow Information
Export Information Elements
(http://www.iana.org/assignments/ipfix/ipfix.xml)".
7.2. Informative References
[I-D.ietf-pmol-metrics-framework]
Clark, A. and B. Claise, "Guidelines for Considering New
Performance Metric Development",
draft-ietf-pmol-metrics-framework-12 (work in progress),
July 2011.
Akhter Expires September 28, 2012 [Page 18]
Internet-Draft PerfMon Meth March 2012
[RFC2508] Casner, S. and V. Jacobson, "Compressing IP/UDP/RTP
Headers for Low-Speed Serial Links", RFC 2508,
February 1999.
[RFC3711] Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K.
Norrman, "The Secure Real-time Transport Protocol (SRTP)",
RFC 3711, March 2004.
[RFC2250] Hoffman, D., Fernando, G., Goyal, V., and M. Civanlar,
"RTP Payload Format for MPEG1/MPEG2 Video", RFC 2250,
January 1998.
[RFC2890] Dommety, G., "Key and Sequence Number Extensions to GRE",
RFC 2890, September 2000.
[RFC4303] Kent, S., "IP Encapsulating Security Payload (ESP)",
RFC 4303, December 2005.
[RFC5761] Perkins, C. and M. Westerlund, "Multiplexing RTP Data and
Control Packets on a Single Port", RFC 5761, April 2010.
[I-D.huici-ipfix-sipfix]
Huici, F., Niccolini, S., and S. Anderson, "SIPFIX: Use
Cases and Problem Statement for VoIP Monitoring and
Exporting", draft-huici-ipfix-sipfix-00 (work in
progress), June 2009.
[nProbe] "nProbe - NetFlow/IPFIX Network Probe
(http://www.ntop.org/nProbe.html)".
[RFC2321] Bressen, A., "RITA -- The Reliable Internetwork
Troubleshooting Agent", RFC 2321, April 1998.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC5477] Dietz, T., Claise, B., Aitken, P., Dressler, F., and G.
Carle, "Information Model for Packet Sampling Exports",
RFC 5477, March 2009.
[VoIP-monitor]
L. Chang-Yong, H. Kim, K. Ko, J. Jim, and H. Jeong, "A
VoIP Traffic Monitoring System based on NetFlow v9,
International Journal of Advanced Science and Technology,
vol. 4, Mar. 2009".
Akhter Expires September 28, 2012 [Page 19]
Internet-Draft PerfMon Meth March 2012
Author's Address
Aamer Akhter
Cisco Systems, Inc.
7025 Kit Creek Road
RTP, NC 27709
USA
Email: aakhter@cisco.com
Akhter Expires September 28, 2012 [Page 20]
