Premier IT Outsourcing and Support Services within the UK

User Tools

Site Tools


Network Working Group S. Floyd, Ed. Request for Comments: 5166 March 2008 Category: Informational

    Metrics for the Evaluation of Congestion Control Mechanisms

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.


 This document is not an IETF Internet Standard.  It represents the
 individual opinion(s) of one or more members of the TMRG Research
 Group of the Internet Research Task Force.  It may be considered for
 standardization by the IETF or adoption as an IRTF Research Group
 consensus document in the future.


 This document discusses the metrics to be considered in an evaluation
 of new or modified congestion control mechanisms for the Internet.
 These include metrics for the evaluation of new transport protocols,
 of proposed modifications to TCP, of application-level congestion
 control, and of Active Queue Management (AQM) mechanisms in the
 router.  This document is the first in a series of documents aimed at
 improving the models that we use in the evaluation of transport
 This document is a product of the Transport Modeling Research Group
 (TMRG), and has received detailed feedback from many members of the
 Research Group (RG).  As the document tries to make clear, there is
 not necessarily a consensus within the research community (or the
 IETF community, the vendor community, the operations community, or
 any other community) about the metrics that congestion control
 mechanisms should be designed to optimize, in terms of trade-offs
 between throughput and delay, fairness between competing flows, and
 the like.  However, we believe that there is a clear consensus that
 congestion control mechanisms should be evaluated in terms of trade-
 offs between a range of metrics, rather than in terms of optimizing
 for a single metric.

Floyd Informational [Page 1] RFC 5166 TMRG, METRICS March 2008

Table of Contents

 1. Introduction ....................................................2
 2. Metrics .........................................................3
    2.1. Throughput, Delay, and Loss Rates ..........................4
         2.1.1. Throughput ..........................................5
         2.1.2. Delay ...............................................6
         2.1.3. Packet Loss Rates ...................................6
    2.2. Response Times and Minimizing Oscillations .................7
         2.2.1. Response to Changes .................................7
         2.2.2. Minimizing Oscillations .............................8
    2.3. Fairness and Convergence ...................................9
         2.3.1. Metrics for Fairness between Flows .................10
         2.3.2. Metrics for Fairness between Flows with
                Different Resource Requirements ....................10
         2.3.3. Comments on Fairness ...............................12
    2.4. Robustness for Challenging Environments ...................13
    2.5. Robustness to Failures and to Misbehaving Users ...........14
    2.6. Deployability .............................................14
    2.7. Metrics for Specific Types of Transport ...................15
    2.8. User-Based Metrics ........................................15
 3. Metrics in the IP Performance Metrics (IPPM) Working Group .....15
 4. Comments on Methodology ........................................16
 5. Security Considerations ........................................16
 6. Acknowledgements ...............................................16
 7. Informative References .........................................17

1. Introduction

 As a step towards improving our methodologies for evaluating
 congestion control mechanisms, in this document we discuss some of
 the metrics to be considered.  We also consider the relationship
 between metrics, e.g., the well-known trade-off between throughput
 and delay.  This document doesn't attempt to specify every metric
 that a study might consider; for example, there are domain-specific
 metrics that researchers might consider that are over and above the
 metrics laid out here.
 We consider metrics for aggregate traffic (taking into account the
 effect of flows on competing traffic in the network) as well as the
 heterogeneous goals of different applications or transport protocols
 (e.g., of high throughput for bulk data transfer, and of low delay
 for interactive voice or video).  Different transport protocols or
 AQM mechanisms might have goals of optimizing different sets of
 metrics, with one transport protocol optimized for per-flow
 throughput and another optimized for robustness over wireless links,
 and with different degrees of attention to fairness with competing
 traffic.  We hope this document will be used as a step in evaluating

Floyd Informational [Page 2] RFC 5166 TMRG, METRICS March 2008

 proposed congestion control mechanisms for a wide range of metrics,
 for example, noting that Mechanism X is good at optimizing Metric A,
 but pays the price with poor performance for Metric B.  The goal
 would be to have a broad view of both the strengths and weaknesses of
 newly proposed congestion control mechanisms.
 Subsequent documents are planned to present sets of simulation and
 testbed scenarios for the evaluation of transport protocols and of
 congestion control mechanisms, based on the best current practice of
 the research community.  These are not intended to be complete or
 final benchmark test suites, but simply to be one step of many to be
 used by researchers in evaluating congestion control mechanisms.
 Subsequent documents are also planned on the methodologies in using
 these sets of scenarios.
 This document is a product of the Transport Modeling Research Group
 (TMRG), and has received detailed feedback from many members of the
 Research Group (RG).  As the document tries to make clear, there is
 not necessarily a consensus within the research community (or the
 IETF community, the vendor community, the operations community, or
 any other community) about the metrics that congestion control
 mechanisms should be designed to optimize, in terms of trade-offs
 between throughput and delay, fairness between competing flows, and
 the like.  However, we believe that there is a clear consensus that
 congestion control mechanisms should be evaluated in terms of
 trade-offs between a range of metrics, rather than in terms of
 optimizing for a single metric.

2. Metrics

 The metrics that we discuss are the following:
 o  Throughput;
 o  Delay;
 o  Packet loss rates;
 o  Response to sudden changes or to transient events;
 o  Minimizing oscillations in throughput or in delay;
 o  Fairness and convergence times;
 o  Robustness for challenging environments;
 o  Robustness to failures and to misbehaving users;

Floyd Informational [Page 3] RFC 5166 TMRG, METRICS March 2008

 o  Deployability;
 o  Metrics for specific types of transport;
 o  User-based metrics.
 We consider each of these below.  Many of the metrics have
 network-based, flow-based, and user-based interpretations.  For
 example, network-based metrics can consider aggregate bandwidth and
 aggregate drop rates, flow-based metrics can consider end-to-end
 transfer times for file transfers or end-to-end delay and packet drop
 rates for interactive traffic, and user-based metrics can consider
 user wait time or user satisfaction with the multimedia experience.
 Our main goal in this document is to explain the set of metrics that
 can be relevant, and not to legislate on the most appropriate
 methodology for using each general metric.
 For some of the metrics, such as fairness, there is not a clear
 agreement in the network community about the desired goals.  In these
 cases, the document attempts to present the range of approaches.

2.1. Throughput, Delay, and Loss Rates

 Because of the clear trade-offs between throughput, delay, and loss
 rates, it can be useful to consider these three metrics together.
 The trade-offs are most clear in terms of queue management at the
 router; is the queue configured to maximize aggregate throughput, to
 minimize delay and loss rates, or some compromise between the two?
 An alternative would be to consider a separate metric such as power,
 defined in this context as throughput over delay, that combines
 throughput and delay.  However, we do not propose in this document a
 clear target in terms of the trade-offs between throughput and delay;
 we are simply proposing that the evaluation of transport protocols
 include an exploration of the competing metrics.
 Using flow-based metrics instead of router-based metrics, the
 relationship between per-flow throughput, delay, and loss rates can
 often be given by the transport protocol itself.  For example, in
 TCP, the sending rate s in packets per second is given as:
    s = 1.2/(RTT*sqrt(p)),
 for the round-trip time RTT and loss rate p [FF99].

Floyd Informational [Page 4] RFC 5166 TMRG, METRICS March 2008

2.1.1. Throughput

 Throughput can be measured as a router-based metric of aggregate link
 utilization, as a flow-based metric of per-connection transfer times,
 and as user-based metrics of utility functions or user wait times.
 It is a clear goal of most congestion control mechanisms to maximize
 throughput, subject to application demand and to the constraints of
 the other metrics.
 Throughput is sometimes distinguished from goodput, where throughput
 is the link utilization or flow rate in bytes per second; goodput,
 also measured in bytes per second, is the subset of throughput
 consisting of useful traffic.  That is, 'goodput' excludes duplicate
 packets, packets that will be dropped downstream, packet fragments or
 ATM cells that are dropped at the receiver because they can't be
 re-assembled into complete packets, and the like.  In general, this
 document doesn't distinguish between throughput and goodput, and uses
 the general term "throughput".
 We note that maximizing throughput is of concern in a wide range of
 environments, from highly-congested networks to under-utilized ones,
 and from long-lived flows to very short ones.  As an example,
 throughput has been used as one of the metrics for evaluating
 Quick-Start, a proposal to allow flows to start up faster than
 slow-start, where throughput has been evaluated in terms of the
 transfer times for connections with a range of transfer sizes
 [RFC4782] [SAF06].
 In some contexts, it might be sufficient to consider the aggregate
 throughput or the mean per-flow throughput [DM06], while in other
 contexts it might be necessary to consider the distribution of
 per-flow throughput.  Some researchers evaluate transport protocols
 in terms of maximizing the aggregate user utility, where a user's
 utility is generally defined as a function of the user's throughput
 Individual applications can have application-specific needs in terms
 of throughput.  For example, real-time video traffic can have highly
 variable bandwidth demands; Voice over IP (VoIP) traffic is sensitive
 to the amount of bandwidth received immediately after idle periods.
 Thus, user metrics for throughput can be more complex than simply the
 per-connection transfer time.

Floyd Informational [Page 5] RFC 5166 TMRG, METRICS March 2008

2.1.2. Delay

 Like throughput, delay can be measured as a router-based metric of
 queueing delay over time, or as a flow-based metric in terms of
 per-packet transfer times.  Per-packet delay can also include delay
 at the sender waiting for the transport protocol to send the packet.
 For reliable transfer, the per-packet transfer time seen by the
 application includes the possible delay of retransmitting a lost
 Users of bulk data transfer applications might care about per-packet
 transfer times only in so far as they affect the per-connection
 transfer time.  On the other end of the spectrum, for users of
 streaming media, per-packet delay can be a significant concern.  Note
 that in some cases the average delay might not capture the metric of
 interest to the users; for example, some users might care about the
 worst-case delay, or about the tail of the delay distribution.
 Note that queueing delay at a router is shared by all flows at a FIFO
 (First-In First-Out) queue.  Thus, the router-based queueing delay
 induced by bulk data transfers can be important even if the bulk data
 transfers themselves are not concerned with per-packet transfer

2.1.3. Packet Loss Rates

 Packet loss rates can be measured as a network-based or as a
 flow-based metric.
 When evaluating the effect of packet losses or ECN marks (Explicit
 Congestion Notification) [RFC3168] on the performance of a congestion
 control mechanism for an individual flow, researchers often use both
 the packet loss/mark rate for that connection and the congestion
 event rate (also called the loss event rate), where a congestion
 event or loss event consists of one or more lost or marked packets in
 one round-trip time [RFC3448].
 Some users might care about the packet loss rate only in so far as it
 affects per-connection transfer times, while other users might care
 about packet loss rates directly.  RFC 3611, RTP Control Protocol
 Extended Reports, describes a VoIP performance-reporting standard
 called RTP Control Protocol Extended Reports (RTCP XR), which
 includes a set of burst metrics.  In RFC 3611, a burst is defined as
 the maximal sequence starting and ending with a lost packet, and not
 including a sequence of Gmin or more packets that are not lost
 [RFC3611].  The burst metrics in RFC 3611 consist of the burst
 density (the fraction of packets in bursts), gap density (the
 fraction of packets in the gaps between bursts), burst duration (the

Floyd Informational [Page 6] RFC 5166 TMRG, METRICS March 2008

 mean duration of bursts in seconds), and gap duration (the mean
 duration of gaps in seconds).  RFC 3357 derives metrics for "loss
 distance" and "loss period", along with statistics that capture loss
 patterns experienced by packet streams on the Internet [RFC3357].
 In some cases, it is useful to distinguish between packets dropped at
 routers due to congestion and packets lost in the network due to
 One network-related reason to avoid high steady-state packet loss
 rates is to avoid congestion collapse in environments containing
 paths with multiple congested links.  In such environments, high
 packet loss rates could result in congested links wasting scarce
 bandwidth by carrying packets that will only be dropped downstream
 before being delivered to the receiver [RFC2914].  We also note that
 in some cases, the retransmit rate can be high, and the goodput
 correspondingly low, even with a low packet drop rate [AEO03].

2.2. Response Times and Minimizing Oscillations

 In this section, we consider response times and oscillations
 together, as there are well-known trade-offs between improving
 response times and minimizing oscillations.  In addition, the
 scenarios that illustrate the dangers of poor response times are
 often quite different from the scenarios that illustrate the dangers
 of unnecessary oscillations.

2.2.1. Response to Changes

 One of the key concerns in the design of congestion control
 mechanisms has been the response times to sudden congestion in the
 network.  On the one hand, congestion control mechanisms should
 respond reasonably promptly to sudden congestion from routing or
 bandwidth changes or from a burst of competing traffic.  At the same
 time, congestion control mechanisms should not respond too severely
 to transient changes, e.g., to a sudden increase in delay that will
 dissipate in less than the connection's round-trip time.
 Congestion control mechanisms also have to contend with sudden
 changes in the bandwidth-delay product due to mobility.  Such
 bandwidth-delay product changes are expected to become more frequent
 and to have greater impact than path changes today.  As a result of
 both mobility and of the heterogeneity of wireless access types
 (802.11b,a,g, WIMAX, WCDMA, HS-WCDMA, E-GPRS, Bluetooth, etc.), both
 the bandwidth and the round-trip delay can change suddenly, sometimes
 by several orders of magnitude.

Floyd Informational [Page 7] RFC 5166 TMRG, METRICS March 2008

 Evaluating the response to sudden or transient changes can be of
 particular concern for slowly responding congestion control
 mechanisms such as equation-based congestion control [RFC3448] and
 AIMD (Additive Increase Multiplicative Decrease) or for related
 mechanisms using parameters that make them more slowly-responding
 than TCP [BB01] [BBFS01].
 In addition to the responsiveness and smoothness of aggregate
 traffic, one can consider the trade-offs between responsiveness,
 smoothness, and aggressiveness for an individual connection [FHP00]
 [YKL01].  In this case, smoothness can be defined by the largest
 reduction in the sending rate in one round-trip time, in a
 deterministic environment with a packet drop exactly every 1/p
 packets.  The responsiveness is defined as the number of round-trip
 times of sustained congestion required for the sender to halve the
 sending rate; aggressiveness is defined as the maximum increase in
 the sending rate in one round-trip time, in packets per second, in
 the absence of congestion.  This aggressiveness can be a function of
 the mode of the transport protocol; for TCP, the aggressiveness of
 slow-start is quite different from the aggressiveness of congestion
 avoidance mode.

2.2.2. Minimizing Oscillations

 One goal is that of stability, in terms of minimizing oscillations of
 queueing delay or of throughput.  In practice, stability is
 frequently associated with rate fluctuations or variance.  Rate
 variations can result in fluctuations in router queue size and
 therefore of queue overflows.  These queue overflows can cause loss
 synchronizations across coexisting flows and periodic
 under-utilization of link capacity, both of which are considered to
 be general signs of network instability.  Thus, measuring the rate
 variations of flows is often used to measure the stability of
 transport protocols.  To measure rate variations, [JWL04], [RX05],
 and [FHPW00] use the coefficient of variation (CoV) of per-flow
 transmission rates, and [WCL05] suggests the use of standard
 deviations of per-flow rates.  Since rate variations are a function
 of time scales, it makes sense to measure these rate variations over
 various time scales.
 Measuring per-flow rate variations, however, is only one aspect of
 transport protocol stability.  A realistic experimental setting
 always involves multiple flows of the transport protocol being
 observed, along with a significant amount of cross traffic, with
 rates varying over time on both the forward and reverse paths.  As a
 congestion control protocol must adapt its rate to the varying rates
 of competing traffic, just measuring the per-flow statistics of a
 subset of the traffic could be misleading because it measures the

Floyd Informational [Page 8] RFC 5166 TMRG, METRICS March 2008

 rate fluctuations due in part to the adaptation to competing traffic
 on the path.  Thus, per-flow statistics are most meaningful if they
 are accompanied by the statistics measured at the network level.  As
 a complementary metric to the per-flow statistics, [HKLRX06] uses
 measurements of the rate variations of the aggregate flows observed
 in bottleneck routers over various time scales.
 Minimizing oscillations in queueing delay or throughput has related
 per-flow metrics of minimizing jitter in round-trip times and loss
 An orthogonal goal for some congestion control mechanisms, e.g., for
 equation-based congestion control, is to minimize the oscillations in
 the sending rate for an individual connection, given an environment
 with a fixed, steady-state packet drop rate.  (As is well known, TCP
 congestion control is characterized by a pronounced oscillation in
 the sending rate, with the sender halving the sending rate in
 response to congestion.)  One metric for the level of oscillations is
 the smoothness metric given in Section 2.2.1 above.
 As discussed in [FK07], the synchronization of loss events can also
 affect convergence times, throughput, and delay.

2.3. Fairness and Convergence

 Another set of metrics is that of fairness and convergence times.
 Fairness can be considered between flows of the same protocol and
 between flows using different protocols (e.g., TCP-friendliness,
 referring to fairness between TCP and a new transport protocol).
 Fairness can also be considered between sessions, between users, or
 between other entities.
 There are a number of different fairness measures.  These include
 max-min fairness [HG86], proportional fairness [KMT98] [K01], the
 fairness index proposed in [JCH84], and the product measure, a
 variant of network power [BJ81].

Floyd Informational [Page 9] RFC 5166 TMRG, METRICS March 2008

2.3.1. Metrics for Fairness between Flows

 This section discusses fairness metrics that consider the fairness
 between flows, but that don't take into account different
 characteristics of flows (e.g., the number of links in the path or
 the round-trip time).  For the discussion of fairness metrics, let
 x_i be the throughput for the i-th connection.
 Jain's fairness index: The fairness index in [JCH84] is:
    (( sum_i x_i )^2) / (n * sum_i ( (x_i)^2 )),
 where there are n users.  This fairness index ranges from 0 to 1, and
 it is maximum when all users receive the same allocation.  This index
 is k/n when k users equally share the resource, and the other n-k
 users receive zero allocation.
 The product measure: The product measure:
    product_i x_i ,
 the product of the throughput of the individual connections, is also
 used as a measure of fairness.  (In some contexts x_i is taken as the
 power of the i-th connection, and the product measure is referred to
 as network power.)  The product measure is particularly sensitive to
 segregation; the product measure is zero if any connection receives
 zero throughput.  In [MS91], it is shown that for a network with many
 connections and one shared gateway, the product measure is maximized
 when all connections receive the same throughput.
 Epsilon-fairness: A rate allocation is defined as epsilon-fair if
    (min_i x_i) / (max_i x_i) >= 1 - epsilon.
 Epsilon-fairness measures the worst-case ratio between any two
 throughput rates [ZKL04].  Epsilon-fairness is related to max-min
 fairness, defined later in this document.

2.3.2. Metrics for Fairness between Flows with Different Resource

 This section discusses metrics for fairness between flows with
 different resource requirements, that is, with different utility
 functions, round-trip times, or number of links on the path.  Many of
 these metrics can be described as solutions to utility maximization
 problems [K01]; the total utility quantifies both the fairness and
 the throughput.

Floyd Informational [Page 10] RFC 5166 TMRG, METRICS March 2008

 Max-min fairness: In order to satisfy the max-min fairness criteria,
 the smallest throughput rate must be as large as possible.  Given
 this condition, the next-smallest throughput rate must be as large as
 possible, and so on.  Thus, the max-min fairness gives absolute
 priority to the smallest flows.  (Max-min fairness can be explained
 by the progressive filling algorithm, where all flow rates start at
 zero, and the rates all grow at the same pace.  Each flow rate stops
 growing only when one or more links on the path reach link capacity.)
 Proportional fairness: In contrast, a feasible allocation, x, is
 defined as proportionally fair if, for any other feasible allocation
 x*, the aggregate of proportional changes is zero or negative:
    sum_i ( (x*_i - x_i)/x_i ) <= 0.
 "This criterion favours smaller flows, but less emphatically than
 max-min fairness" [K01].  (Using the language of utility functions,
 proportional fairness can be achieved by using logarithmic utility
 functions, and maximizing the sum of the per-flow utility functions;
 see [KMT98] for a fuller explanation.)
 Minimum potential delay fairness: Minimum potential delay fairness
 has been shown to model TCP [KS03], and is a compromise between
 max-min fairness and proportional fairness.  An allocation, x, is
 defined as having minimum potential delay fairness if:
    sum_i (1/x_i)
 is smaller than for any other feasible allocation.  That is, it would
 minimize the average download time if each flow was an equal-sized
 In [CRM05], Colussi, et al. propose a new definition of TCP fairness,
 that "a set of TCP fair flows do not cause more congestion than a set
 of TCP flows would cause", where congestion is defined in terms of
 queueing delay, queueing delay variation, the congestion event rate
 [e.g., loss event rate], and the packet loss rate.
 Chiu and Tan in [CT06] argue for redefining the notion of fairness
 when studying traffic controls for inelastic traffic, proposing that
 inelastic flows adopt other traffic controls such as admission
 The usefulness of flow-rate fairness has been challenged in [B07] by
 Briscoe, and defended in [FA08] by Floyd and Allman.  In [B07],
 Briscoe argues that flow-rate fairness should not be a desired goal,
 and that instead "we should judge fairness mechanisms on how they
 share out the 'cost' of each user's actions on others".  Floyd and

Floyd Informational [Page 11] RFC 5166 TMRG, METRICS March 2008

 Allman in [FA08] argue that the current system based on TCP
 congestion control and flow-rate fairness has been useful in the real
 world, posing minimal demands on network and economic infrastructure
 and enabling users to get a share of the network resources.

2.3.3. Comments on Fairness

 Trade-offs between fairness and throughput: The fairness measures in
 the section above generally measure both fairness and throughput,
 giving different weights to each.  Potential trade-offs between
 fairness and throughput are also discussed by Tang, et al. in
 [TWL06], for a framework where max-min fairness is defined as the
 most fair.  In particular, [TWL06] shows that in some topologies,
 throughput is proportional to fairness, while in other topologies,
 throughput is inversely proportional to fairness.
 Fairness and the number of congested links: Some of these fairness
 metrics are discussed in more detail in [F91].  We note that there is
 not a clear consensus for the fairness goals, in particular for
 fairness between flows that traverse different numbers of congested
 links [F91].  Utility maximization provides one framework for
 describing this trade-off in fairness.
 Fairness and round-trip times: One goal cited in a number of new
 transport protocols has been that of fairness between flows with
 different round-trip times [KHR02] [XHR04].  We note that there is
 not a consensus in the networking community about the desirability of
 this goal, or about the implications and interactions between this
 goal and other metrics [FJ92] (Section 3.3).  One common argument
 against the goal of fairness between flows with different round-trip
 times has been that flows with long round-trip times consume more
 resources; this aspect is covered by the previous paragraph.
 Researchers have also noted the difference between the RTT-unfairness
 of standard TCP, and the greater RTT-unfairness of some proposed
 modifications to TCP [LLS05].
 Fairness and packet size: One fairness issue is that of the relative
 fairness for flows with different packet sizes.  Many file transfer
 applications will use the maximum packet size possible;  in contrast,
 low-bandwidth VoIP flows are likely to send small packets, sending a
 new packet every 10 to 40 ms., to limit delay.  Should a small-packet
 VoIP connection receive the same sending rate in *bytes* per second
 as a large-packet TCP connection in the same environment, or should
 it receive the same sending rate in *packets* per second?  This
 fairness issue has been discussed in more detail in [RFC3714], with
 [RFC4828] also describing the ways that packet size can affect the
 packet drop rate experienced by a flow.

Floyd Informational [Page 12] RFC 5166 TMRG, METRICS March 2008

 Convergence times: Convergence times concern the time for convergence
 to fairness between an existing flow and a newly starting one, and
 are a special concern for environments with high-bandwidth long-delay
 flows.  Convergence times also concern the time for convergence to
 fairness after a sudden change such as a change in the network path,
 the competing cross-traffic, or the characteristics of a wireless
 link.  As with fairness, convergence times can matter both between
 flows of the same protocol, and between flows using different
 protocols [SLFK03].  One metric used for convergence times is the
 delta-fair convergence time, defined as the time taken for two flows
 with the same round-trip time to go from shares of 100/101-th and
 1/101-th of the link bandwidth, to having close to fair sharing with
 shares of (1+delta)/2 and (1-delta)/2 of the link bandwidth [BBFS01].
 A similar metric for convergence times measures the convergence time
 as the number of round-trip times for two flows to reach epsilon-
 fairness, when starting from a maximally-unfair state [ZKL04].

2.4. Robustness for Challenging Environments

 While congestion control mechanisms are generally evaluated first
 over environments with static routing in a network of two-way
 point-to-point links, some environments bring up more challenging
 problems (e.g., corrupted packets, reordering, variable bandwidth,
 and mobility) as well as new metrics to be considered (e.g., energy
 Robustness for challenging environments: Robustness needs to be
 explored for paths with reordering, corruption, variable bandwidth,
 asymmetric routing, router configuration changes, mobility, and the
 like [GF04].  In general, the Internet architecture has valued
 robustness over efficiency, e.g., when there are trade-offs between
 robustness and the throughput, delay, and fairness metrics described
 Energy consumption: In mobile environments, the energy consumption
 for the mobile end-node can be a key metric that is affected by the
 transport protocol [TM02].
 The goodput ratio: For wireless networks, the goodput ratio can be a
 useful metric, where the goodput ratio can be defined as the useful
 data delivered to users as a fraction of the total amount of data
 transmitted on the network.  A high goodput ratio indicates an
 efficient use of the radio spectrum and lower interference with other

Floyd Informational [Page 13] RFC 5166 TMRG, METRICS March 2008

2.5. Robustness to Failures and to Misbehaving Users

 One goal is for congestion control mechanisms to be robust to
 misbehaving users, such as receivers that 'lie' to data senders about
 the congestion experienced along the path or otherwise attempt to
 bypass the congestion control mechanisms of the sender [SCWA99].
 Another goal is for congestion control mechanisms to be as robust as
 possible to failures, such as failures of routers in using explicit
 feedback to end-nodes or failures of end-nodes to follow the
 prescribed protocols.

2.6. Deployability

 One metric for congestion control mechanisms is their deployability
 in the current Internet.  Metrics related to deployability include
 the ease of failure diagnosis and the overhead in terms of packet
 header size or added complexity at end-nodes or routers.
 One key aspect of deployability concerns the range of deployment
 needed for a new congestion control mechanism.  Consider the
 following possible deployment requirements:
  • Only at the sender (e.g., NewReno in TCP [RFC3782]);
  • Only at the receiver (e.g., delayed acknowledgements in TCP);
  • Both the sender and receiver (e.g., Selective Acknowledgment

(SACK) TCP [RFC2018]);

  • At a single router (e.g., Random Early Detection (RED) [FJ93]);
  • All of the routers along the end-to-end path;
  • Both end-nodes and all routers along the path (e.g., Explicit

Control Protocol (XCP) [KHR02]).

 Some congestion control mechanisms (e.g., XCP [KHR02], Quick-Start
 [RFC4782]) may also have deployment issues with IPsec, IP in IP,
 MPLS, other tunnels, or with non-router queues such as those in
 Ethernet switches.

Floyd Informational [Page 14] RFC 5166 TMRG, METRICS March 2008

 Another deployability issue concerns the complexity of the code.  How
 complex is the code required to implement the mechanism in software?
 Is floating point math required?  How much new state must be kept to
 implement the new mechanism, and who holds that state, the routers or
 the end-nodes?  We note that we don't suggest these questions as ways
 to reduce the deployability metric to a single number; we suggest
 them as issues that could be considered in evaluating the
 deployability of a proposed congestion control mechanism.

2.7. Metrics for Specific Types of Transport

 In some cases, modified metrics are needed for evaluating transport
 protocols intended for quality-of-service (QoS)-enabled environments
 or for below-best-effort traffic [VKD02] [KK03].

2.8. User-Based Metrics

 An alternate approach that has been proposed for the evaluation of
 congestion control mechanisms would be to evaluate in terms of user
 metrics, such as user satisfaction or in terms of
 application-specific utility functions.  Such an approach would
 require the definition of a range of user metrics or of
 application-specific utility functions for the range of applications
 under consideration (e.g., FTP, HTTP, VoIP).

3. Metrics in the IP Performance Metrics (IPPM) Working Group

 The IPPM Working Group [IPPM] was established to define performance
 metrics to be used by network operators, end users, or independent
 testing groups.  The metrics include metrics for connectivity
 [RFC2678], delay and loss [RFC2679], [RFC2680], and [RFC2681], delay
 variation [RFC3393], loss patterns [RFC3357], packet reordering
 [RFC4737], bulk transfer capacity [RFC3148], and link capacity
 [RFC5136].  The IPPM documents give concrete, well-defined metrics,
 along with a methodology for measuring the metric.  The metrics
 discussed in this document have a different purpose from the IPPM
 metrics; in this document, we are discussing metrics as used in
 analysis, simulations, and experiments for the evaluation of
 congestion control mechanisms.  Further, we are discussing these
 metrics in a general sense, rather than looking for specific concrete
 definitions for each metric.  However, there are many cases where the
 metric definitions from IPPM could be useful, for specific issues of
 how to measure these metrics in simulations, or in testbeds, and for
 providing common definitions for talking about these metrics.

Floyd Informational [Page 15] RFC 5166 TMRG, METRICS March 2008

4. Comments on Methodology

 The types of scenarios that are used to test specific metrics, and
 the range of parameters that it is useful to consider, will be
 discussed in separate documents, e.g., along with specific scenarios
 for use in evaluating congestion control mechanisms.
 We note that it can be important to evaluate metrics over a wide
 range of environments, with a range of link bandwidths, congestion
 levels, and levels of statistical multiplexing.  It is also important
 to evaluate congestion control mechanisms in a range of scenarios,
 including typical ranges of connection sizes and round-trip times
 [FK02].  It is also useful to compare metrics for new or modified
 transport protocols with those of the current standards for TCP.
 As an example from the literature on evaluating transport protocols,
 Li, et al. in "Experimental Evaluation of TCP Protocols for High-
 Speed Networks" [LLS05] focus on the performance of TCP in high-
 speed networks, and consider metrics for aggregate throughput, loss
 rates, fairness (including fairness between flows with different
 round-trip times), response times (including convergence times), and
 incremental deployment.  More general references on methodology
 include [J91]. Papers that discuss the range of metrics for
 evaluating congestion control include [MTZ04].

5. Security Considerations

 Section 2.5 discusses the robustness of congestion control mechanisms
 to failures and to misbehaving users.  Transport protocols also have
 other security concerns that are unrelated to congestion control
 mechanisms; these are not discussed in this document.

6. Acknowledgements

 Thanks to Lachlan Andrew, Mark Allman, Armando Caro, Dah Ming Chiu,
 Eric Coe, Dado Colussi, Wesley Eddy, Aaron Falk, Nelson Fonseca,
 Janardhan Iyengar, Doug Leith, Sara Landstrom, Tony Li, Saverio
 Mascolo, Sean Moore, Injong Rhee, David Ros, Juergen Schoenwaelder,
 Andras Veres, Michael Welzl, and Damon Wischik, and members of the
 Transport Modeling Research Group for feedback and contributions.

Floyd Informational [Page 16] RFC 5166 TMRG, METRICS March 2008

7. Informative References

 [AEO03]   M. Allman, W. Eddy, and S. Ostermann, Estimating Loss Rates
           With TCP, ACM Performance Evaluation Review, 31(3),
           December 2003.
 [BB01]    D. Bansal and H. Balakrishnan, Binomial Congestion Control
           Algorithms, IEEE Infocom, April 2001.
 [BBFS01]  D. Bansal, H. Balakrishnan, S. Floyd, and S. Shenker,
           Dynamic Behavior of Slowly-Responsive Congestion Control
           Algorithms, SIGCOMM 2001.
 [BJ81]    K. Bharath-Kumar and J. Jeffrey, A New Approach to
           Performance-Oriented Flow Control, IEEE Transactions on
           Communications, Vol.29 N.4, April 1981.
 [B07]     B. Briscoe, "Flow Rate Fairness: Dismantling a Religion",
           Computer Communications Review, V.37 N.2, April 2007.
 [CRM05]   D. Colussi, A New Approach to TCP-Fairness, Report C-2005-
           49, University of Helsinki, Finland, 2005.
 [CT06] D. Chiu and A. Tam, Redefining Fairness in the Study of
           TCP-friendly Traffic Controls, Technical Report, 2006.
 [DM06]    N. Dukkipati and N. McKeown, Why Flow-Completion Time is
           the Right Metric for Congestion Control, ACM SIGCOMM,
           January 2006.
 [F91]     S. Floyd, Connections with Multiple Congested Gateways in
           Packet-Switched Networks Part 1: One-way Traffic, Computer
           Communication Review, Vol.21 No.5, October 1991, p. 30-47.
 [FA08]    S. Floyd and M. Allman, Comments on the Usefulness of
           Simple Best-Effort Traffic, Work in Progress, January 2007.
 [FF99]    Floyd, S., and Fall, K., "Promoting the Use of End-to-End
           Congestion Control in the Internet", IEEE/ACM Transactions
           on Networking, August 1999.
 [FHP00]   S. Floyd, M. Handley, and J. Padhye, A Comparison of
           Equation-Based and AIMD Congestion Control, May 2000.   URL
 [FHPW00]  S. Floyd, M. Handley, J. Padhye, and J. Widmer, Equation-
           Based Congestion Control for Unicast Applications, SIGCOMM
           2000, August 2000.

Floyd Informational [Page 17] RFC 5166 TMRG, METRICS March 2008

 [FJ92]    S. Floyd and V. Jacobson, On Traffic Phase Effects in
           Packet-Switched Gateways, Internetworking: Research and
           Experience, V.3 N.3, September 1992, p.115-156.
 [FJ93]    S. Floyd and V. Jacobson, Random Early Detection gateways
           for Congestion Avoidance, IEEE/ACM Transactions on
           Networking, V.1 N.4, August 1993,
 [FK02]    S. Floyd and E. Kohler, Internet Research Needs Better
           Models, Hotnets-I. October 2002.
 [FK07]    S. Floyd and E. Kohler, "Tools for the Evaluation of
           Simulation and Testbed Scenarios", Work in Progress,
           February 2008.
 [GF04]    A. Gurtov and S. Floyd, Modeling Wireless Links for
           Transport Protocols, ACM CCR, 34(2):85-96, April 2004.
 [HKLRX06] S. Ha, Y. Kim, L. Le, I. Rhee, and L. Xu, A Step Toward
           Realistic Evaluation of High-speed TCP Protocols, technical
           report, North Carolina State University, January 2006.
 [HG86]    E. Hahne and R. Gallager, Round Robin Scheduling for Fair
           Flow Control in Data Communications Networks, IEEE
           International Conference on Communications, June 1986.
 [IPPM]    IP Performance Metrics (IPPM) Working Group, URL
 [J91]     R. Jain, The Art of Computer Systems Performance Analysis:
           Techniques for Experimental Design, Measurement,
           Simulation, and Modeling, John Wiley & Sons, 1991.
 [JCH84]   R. Jain, D.M. Chiu, and W. Hawe, A Quantitative Measure of
           Fairness and Discrimination for Resource Allocation in
           Shared Systems, DEC TR-301, Littleton, MA: Digital
           Equipment Corporation, 1984.
 [JWL04]   C. Jin, D. Wei, and S. Low, FAST TCP: Motivation,
           Architecture, Algorithms, Performance, IEEE INFOCOM, March
 [K01]     F. Kelly, Mathematical Modelling of the Internet,
           "Mathematics Unlimited - 2001 and Beyond" (Editors B.
           Engquist and W.  Schmid), Springer-Verlag, Berlin, pp.
           685-702, 2001.

Floyd Informational [Page 18] RFC 5166 TMRG, METRICS March 2008

 [KHR02]   D. Katabi, M. Handley, and C. Rohrs, Congestion Control for
           High Bandwidth-Delay Product Networks, ACM Sigcomm, 2002.
 [KK03]    A. Kuzmanovic and E. W. Knightly, TCP-LP: A Distributed
           Algorithm for Low Priority Data Transfer, IEEE INFOCOM
           2003, April 2003.
 [KMT98]   F. Kelly, A. Maulloo and D. Tan, Rate Control in
           Communication Networks: Shadow Prices, Proportional
           Fairness and Stability.  Journal of the Operational
           Research Society 49, pp. 237-252, 1998.
 [KS03]    S. Kunniyur and R. Srikant, End-to-end Congestion Control
           Schemes: Utility Functions, Random Losses and ECN Marks,
           IEEE/ACM Transactions on Networking, 11(5):689-702, October
 [LLS05]   Y-T. Li, D. Leith, and R. Shorten, Experimental Evaluation
           of TCP Protocols for High-Speed Networks, Hamilton
           Institute, 2005.  URL
 [MS91]    D. Mitra and J. Seery, Dynamic Adaptive Windows for High
           Speed Data Networks with Multiple Paths and Propagation
           Delays, INFOCOM '91, pp 39-48.
 [MTZ04]   L. Mamatas, V. Tsaoussidis, and C. Zhang, Approaches to
           Congestion Control in Packet Networks, 2004.
 [RFC2018] Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP
           Selective Acknowledgment Options", RFC 2018, October 1996.
 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way
           Packet Loss Metric for IPPM", RFC 2680, September 1999.
 [RFC2678] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring
           Connectivity", RFC 2678, September 1999.
 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way
           Delay Metric for IPPM", RFC 2679, September 1999.
 [RFC2681] Almes, G., Kalidindi, S., and M. Zekauskas, "A Round-trip
           Delay Metric for IPPM", RFC 2681, September 1999.
 [RFC2914] Floyd, S., "Congestion Control Principles", BCP 41, RFC
           2914, September 2000.

Floyd Informational [Page 19] RFC 5166 TMRG, METRICS March 2008

 [RFC3148] Mathis, M. and M. Allman, "A Framework for Defining
           Empirical Bulk Transfer Capacity Metrics", RFC 3148, July
 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition of
           Explicit Congestion Notification (ECN) to IP", RFC 3168,
           September 2001.
 [RFC3357] Koodli, R. and R. Ravikanth, "One-way Loss Pattern Sample
           Metrics", RFC 3357, August 2002.
 [RFC3393] Demichelis, C. and P. Chimento, "IP Packet Delay Variation
           Metric for IP Performance Metrics (IPPM)", RFC 3393,
           November 2002.
 [RFC3448] Handley, M., Floyd, S., Padhye, J., and J. Widmer, "TCP
           Friendly Rate Control (TFRC): Protocol Specification", RFC
           3448, January 2003.
 [RFC3611] Friedman, T., Ed., Caceres, R., Ed., and A. Clark, Ed.,
           "RTP Control Protocol Extended Reports (RTCP XR)", RFC
           3611, November 2003.
 [RFC3714] Floyd, S., Ed., and J. Kempf, Ed., "IAB Concerns Regarding
           Congestion Control for Voice Traffic in the Internet", RFC
           3714, March 2004.
 [RFC3782] Floyd, S., Henderson, T., and A. Gurtov, "The NewReno
           Modification to TCP's Fast Recovery Algorithm", RFC 3782,
           April 2004.
 [RFC4737] Morton, A., Ciavattone, L., Ramachandran, G., Shalunov, S.,
           and J. Perser, "Packet Reordering Metrics", RFC 4737,
           November 2006.
 [RFC4782] Floyd, S., Allman, M., Jain, A., and P. Sarolahti, "Quick-
           Start for TCP and IP", RFC 4782, January 2007.
 [RFC4828] Floyd, S. and E. Kohler, "TCP Friendly Rate Control (TFRC):
           The Small-Packet (SP) Variant", RFC 4828, April 2007.
 [RFC5136] Chimento, P. and J. Ishac, "Defining Network Capacity", RFC
           5136, February 2008.
 [RX05]    I. Rhee and L. Xu, CUBIC: A New TCP-Friendly High-Speed TCP
           Variant, PFLDnet 2005.

Floyd Informational [Page 20] RFC 5166 TMRG, METRICS March 2008

 [SAF06]   P. Sarolahti, M. Allman, and S. Floyd, Determining an
           Appropriate Sending Rate Over an Underutilized Network
           Path, Computer Networks, September 2006.
 [SLFK03]  R.N. Shorten, D.J. Leith, J. Foy, and R. Kilduff, Analysis
           and Design of Congestion Control in Synchronised
           Communication Networks. Proc. 12th Yale Workshop on
           Adaptive & Learning Systems, May 2003.
 [SCWA99]  S. Savage, N. Cardwell, D. Wetherall, and T. Anderson, TCP
           Congestion Control with a Misbehaving Receiver, ACM
           Computer Communications Review, October 1999.
 [TM02]    V. Tsaoussidis and I. Matta, Open Issues of TCP for Mobile
           Computing, Journal of Wireless Communications and Mobile
           Computing: Special Issue on Reliable Transport Protocols
           for Mobile Computing, February 2002.
 [TWL06]   A. Tang, J. Wang and S. H. Low.  Counter-Intuitive
           Throughput Behaviors in Networks Under End-to-End Control,
           IEEE/ACM Transactions on Networking, 14(2):355-368, April
 [WCL05]   D. X. Wei, P. Cao and S. H. Low, Time for a TCP Benchmark
           Suite?, Technical Report, Caltech CS, Stanford EAS,
           Caltech, 2005.
 [VKD02]   A. Venkataramani, R. Kokku, and M. Dahlin, TCP Nice: A
           Mechanism for Background Transfers, Fifth USENIX Symposium
           on Operating System Design and Implementation (OSDI), 2002.
 [XHR04]   L. Xu, K. Harfoush, and I. Rhee, Binary Increase Congestion
           Control for Fast, Long Distance Networks, Infocom 2004.
 [YKL01]   Y. Yang, M. Kim, and S. Lam, Transient Behaviors of TCP-
           friendly Congestion Control Protocols, Infocom 2001.
 [ZKL04]   Y. Zhang, S.-R. Kang, and D. Loguinov, Delayed Stability
           and Performance of Distributed Congestion Control, ACM
           SIGCOMM, August 2004.

Floyd Informational [Page 21] RFC 5166 TMRG, METRICS March 2008

Author's Address

 Sally Floyd
 ICSI Center for Internet Research
 1947 Center Street, Suite 600
 Berkeley, CA 94704

Floyd Informational [Page 22] RFC 5166 TMRG, METRICS March 2008

Full Copyright Statement

 Copyright (C) The IETF Trust (2008).
 This document is subject to the rights, licenses and restrictions
 contained in BCP 78 and at,
 and except as set forth therein, the authors retain all their rights.
 This document and the information contained herein are provided on an

Intellectual Property

 The IETF takes no position regarding the validity or scope of any
 Intellectual Property Rights or other rights that might be claimed to
 pertain to the implementation or use of the technology described in
 this document or the extent to which any license under such rights
 might or might not be available; nor does it represent that it has
 made any independent effort to identify any such rights.  Information
 on the procedures with respect to rights in RFC documents can be
 found in BCP 78 and BCP 79.
 Copies of IPR disclosures made to the IETF Secretariat and any
 assurances of licenses to be made available, or the result of an
 attempt made to obtain a general license or permission for the use of
 such proprietary rights by implementers or users of this
 specification can be obtained from the IETF on-line IPR repository at
 The IETF invites any interested party to bring to its attention any
 copyrights, patents or patent applications, or other proprietary
 rights that may cover technology that may be required to implement
 this standard.  Please address the information to the IETF at

Floyd Informational [Page 23]

/data/webs/external/dokuwiki/data/pages/rfc/rfc5166.txt · Last modified: 2008/03/18 22:50 by

Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki