Internet DRAFT - draft-ietf-aqm-fq-codel

draft-ietf-aqm-fq-codel







AQM working group                                 T. Hoeiland-Joergensen
Internet-Draft                                       Karlstad University
Intended status: Experimental                                P. McKenney
Expires: September 19, 2016                  IBM Linux Technology Center
                                                                 D. Taht
                                                                Teklibre
                                                               J. Gettys

                                                              E. Dumazet
                                                            Google, Inc.
                                                          March 18, 2016


    The FlowQueue-CoDel Packet Scheduler and Active Queue Management
                               Algorithm
                       draft-ietf-aqm-fq-codel-06

Abstract

   This memo presents the FQ-CoDel hybrid packet scheduler/Active Queue
   Management algorithm, a powerful tool for fighting bufferbloat and
   reducing latency.

   FQ-CoDel mixes packets from multiple flows and reduces the impact of
   head of line blocking from bursty traffic.  It provides isolation for
   low-rate traffic such as DNS, web, and videoconferencing traffic.  It
   improves utilisation across the networking fabric, especially for
   bidirectional traffic, by keeping queue lengths short; and it can be
   implemented in a memory- and CPU-efficient fashion across a wide
   range of hardware.

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 19, 2016.




Hoeiland-Joergensen, etExpires September 19, 2016               [Page 1]

Internet-Draft                  fq-codel                      March 2016


Copyright Notice

   Copyright (c) 2016 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
   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
     1.1.  Conventions used in this document . . . . . . . . . . . .   4
     1.2.  Terminology and concepts  . . . . . . . . . . . . . . . .   4
     1.3.  Informal summary of FQ-CoDel  . . . . . . . . . . . . . .   5
   2.  CoDel . . . . . . . . . . . . . . . . . . . . . . . . . . . .   6
   3.  Flow Queueing . . . . . . . . . . . . . . . . . . . . . . . .   6
   4.  The FQ-CoDel scheduler  . . . . . . . . . . . . . . . . . . .   7
     4.1.  Enqueue . . . . . . . . . . . . . . . . . . . . . . . . .   7
       4.1.1.  Alternative classification schemes  . . . . . . . . .   8
     4.2.  Dequeue . . . . . . . . . . . . . . . . . . . . . . . . .   9
   5.  Implementation considerations . . . . . . . . . . . . . . . .  10
     5.1.  Data structures . . . . . . . . . . . . . . . . . . . . .  10
     5.2.  Parameters  . . . . . . . . . . . . . . . . . . . . . . .  11
       5.2.1.  Interval  . . . . . . . . . . . . . . . . . . . . . .  11
       5.2.2.  Target  . . . . . . . . . . . . . . . . . . . . . . .  11
       5.2.3.  Packet limit  . . . . . . . . . . . . . . . . . . . .  11
       5.2.4.  Quantum . . . . . . . . . . . . . . . . . . . . . . .  12
       5.2.5.  Flows . . . . . . . . . . . . . . . . . . . . . . . .  12
       5.2.6.  Explicit Congestion Notification (ECN)  . . . . . . .  12
       5.2.7.  CE threshold  . . . . . . . . . . . . . . . . . . . .  13
     5.3.  Probability of hash collisions  . . . . . . . . . . . . .  13
     5.4.  Memory Overhead . . . . . . . . . . . . . . . . . . . . .  13
     5.5.  Per-Packet Timestamping . . . . . . . . . . . . . . . . .  14
     5.6.  Limiting queueing in lower layers . . . . . . . . . . . .  15
     5.7.  Other forms of "Fair Queueing"  . . . . . . . . . . . . .  15
     5.8.  Differences between CoDel and FQ-CoDel behaviour  . . . .  15
   6.  Limitations of flow queueing  . . . . . . . . . . . . . . . .  16
     6.1.  Fairness between things other than flows  . . . . . . . .  16
     6.2.  Flow bunching by opaque encapsulation . . . . . . . . . .  17
     6.3.  Low-priority congestion control algorithms  . . . . . . .  17
   7.  Deployment status and future work . . . . . . . . . . . . . .  18



Hoeiland-Joergensen, etExpires September 19, 2016               [Page 2]

Internet-Draft                  fq-codel                      March 2016


   8.  Security Considerations . . . . . . . . . . . . . . . . . . .  18
   9.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  19
   10. Acknowledgements  . . . . . . . . . . . . . . . . . . . . . .  19
   11. References  . . . . . . . . . . . . . . . . . . . . . . . . .  19
     11.1.  Normative References . . . . . . . . . . . . . . . . . .  19
     11.2.  Informative References . . . . . . . . . . . . . . . . .  21
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  22

1.  Introduction

   The FlowQueue-CoDel (FQ-CoDel) algorithm is a combined packet
   scheduler and Active Queue Management (AQM) [RFC3168] algorithm
   developed as part of the bufferbloat-fighting community effort
   [BLOATWEB].  It is based on a modified Deficit Round Robin (DRR)
   queue scheduler [DRR][DRRPP], with the CoDel AQM [I-D.ietf-aqm-codel]
   algorithm operating on each queue.  This document describes the
   combined algorithm; reference implementations are available for the
   ns2 [NS2] and ns3 [NS3] network simulators, and it is included in the
   mainline Linux kernel as the fq_codel queueing discipline [LINUXSRC].

   FQ-CoDel is a general, efficient, nearly parameterless queue
   management approach combining flow queueing with CoDel.  It is a
   powerful tool for solving bufferbloat [BLOAT], and we believe it to
   be safe to turn on by default, as has already happened in a number of
   Linux distributions.  In this document we document the Linux
   implementation in sufficient detail for an independent
   implementation, to enable deployment outside of the Linux ecosystem.

   Since the FQ-CoDel algorithm was originally developed in the Linux
   kernel, that implementation is still considered canonical.  This
   document strives to describe the algorithm in the abstract in the
   first sections and separate out most implementation details in
   subsequent sections, but does use the Linux implementation as
   reference for default behaviour in the algorithm description itself.

   The rest of this document is structured as follows: This section
   gives some concepts and terminology used in the rest of the document,
   and gives a short informal summary of the FQ-CoDel algorithm.
   Section 2 gives an overview of the CoDel algorithm.  Section 3 covers
   the flow hashing and DRR portion.  Section 4 then describes the
   working of the algorithm in detail, while Section 5 describes
   implementation details and considerations.  Section 6 lists some of
   the limitations of using flow queueing.  Finally, Section 7 outlines
   the current status of FQ-CoDel deployment and lists some possible
   future areas of inquiry, and Section 8 reiterates some important
   security points that must be observed in the implementation.





Hoeiland-Joergensen, etExpires September 19, 2016               [Page 3]

Internet-Draft                  fq-codel                      March 2016


1.1.  Conventions used in this document

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in [RFC2119].

   In this document, these words will appear with that interpretation
   only when in ALL CAPS.  Lower case uses of these words are not to be
   interpreted as carrying [RFC2119] significance.

1.2.  Terminology and concepts

   Flow: A flow is typically identified by a 5-tuple of source IP,
   destination IP, source port, destination port, and protocol number.
   It can also be identified by a superset or subset of those
   parameters, or by media access control (MAC) address, or other means.
   FQ-CoDel hashes flows into a configurable number of buckets to assign
   packets to internal Queues.

   Queue: A queue of packets represented internally in FQ-CoDel.  In
   most instances each flow gets its own queue; however because of the
   possibility of hash collisions, this is not always the case.  In an
   attempt to avoid confusion, the word 'queue' is used to refer to the
   internal data structure, and 'flow' to refer to the actual stream of
   packets being delivered to the FQ-CoDel algorithm.

   Scheduler: A mechanism to select which queue a packet is dequeued
   from.

   CoDel AQM: The Active Queue Management algorithm employed by FQ-CoDel
   [I-D.ietf-aqm-codel].

   DRR: Deficit round-robin scheduling [DRR].

   Quantum: The maximum amount of bytes to be dequeued from a queue at
   once.

   Interval: Characteristic time period used by the control loop of
   CoDel to detect when a persistent Queue is developing (see
   Section 4.3 of [I-D.ietf-aqm-codel]).

   Target: Setpoint value of the minimum sojourn time of packets in a
   Queue used as the target of the control loop in CoDel (see
   Section 4.4 of [I-D.ietf-aqm-codel]).







Hoeiland-Joergensen, etExpires September 19, 2016               [Page 4]

Internet-Draft                  fq-codel                      March 2016


1.3.  Informal summary of FQ-CoDel

   FQ-CoDel is a _hybrid_ of DRR [DRR] and CoDel [I-D.ietf-aqm-codel],
   with an optimisation for sparse flows similar to Shortest Queue First
   (SQF) [SQF] and DRR++ [DRRPP].  We call this "Flow Queueing" rather
   than "Fair Queueing" as flows that build a queue are treated
   differently from flows that do not.

   By default, FQ-CoDel stochastically classifies incoming packets into
   different queues by hashing the 5-tuple of IP protocol number and
   source and destination IP and port numbers, perturbed with a random
   number selected at initiation time (although other flow
   classification schemes can optionally be configured instead; see
   Section 4.1.1).  Each queue is managed by the CoDel AQM algorithm
   [CODEL].  Packet ordering within a queue is preserved, since queues
   have FIFO ordering.

   The FQ-CoDel algorithm consists of two logical parts: the scheduler
   which selects which queue to dequeue a packet from, and the CoDel AQM
   which works on each of the queues.  The subtleties of FQ-CoDel are
   mostly in the scheduling part, whereas the interaction between the
   scheduler and the CoDel algorithm are fairly straight forward:

   At initialisation, each queue is set up to have a separate set of
   CoDel state variables.  By default, 1024 queues are created.  The
   Linux implementation at the time of writing supports anywhere from
   one to 64K separate queues, and each queue maintains the state
   variables throughout its lifetime, and so acts the same as the non-FQ
   CoDel variant would.  This means that with only one queue, FQ-CoDel
   behaves essentially the same as CoDel by itself.

   On dequeue, FQ-CoDel selects a queue from which to dequeue by a two-
   tier round-robin scheme, in which each queue is allowed to dequeue up
   to a configurable quantum of bytes for each iteration.  Deviations
   from this quantum is maintained as byte credits for the queue, which
   serves to make the fairness scheme byte-based rather than packet-
   based.  The two-tier round-robin mechanism distinguishes between
   "new" queues (which don't build up a standing queue) and "old"
   queues, that have queued enough data to be around for more than one
   iteration of the round-robin scheduler.

   This new/old queue distinction has a particular consequence for
   queues that don't build up more than a quantum of bytes before being
   visited by the scheduler: Such queues are removed from the list, and
   then re-added as a new queue each time a packet arrives for it, and
   so will get priority over queues that do not empty out each round
   (except for a minor modification to protect against starvation,
   detailed below).  Exactly how little data a flow has to send to keep



Hoeiland-Joergensen, etExpires September 19, 2016               [Page 5]

Internet-Draft                  fq-codel                      March 2016


   its queue in this state is somewhat difficult to reason about,
   because it depends on both the egress link speed and the number of
   concurrent flows.  However, in practice many things that are
   beneficial to have prioritised for typical internet use (ACKs, DNS
   lookups, interactive SSH, HTTP requests, VoIP) _tend_ to fall in this
   category, which is why FQ-CoDel performs so well for many practical
   applications.  However, the implicitness of the prioritisation means
   that for applications that require guaranteed priority (for instance
   multiplexing the network control plane over the network itself),
   explicit classification is still needed.

   This scheduling scheme has some subtlety to it, which is explained in
   detail in the remainder of this document.

2.  CoDel

   CoDel is described in the ACM Queue paper [CODEL], and the IETF
   document [I-D.ietf-aqm-codel].  The basic idea is to control queue
   length, maintaining sufficient queueing to keep the outgoing link
   busy, but avoiding building up the queue beyond that point.  This is
   done by preferentially dropping packets that remain in the queue for
   "too long".  Packets are dropped by head drop, which lowers the time
   for the drop signal to propagate back to the sender by the length of
   the queue, and helps trigger TCP fast retransmit sooner.

   The CoDel algorithm itself will not be described here; instead we
   refer the reader to the CoDel draft [I-D.ietf-aqm-codel].

3.  Flow Queueing

   The intention of FQ-CoDel's scheduler is to give each _flow_ its own
   queue, hence the term _Flow Queueing_. Rather than a perfect
   realisation of this, a hashing-based scheme is used, where flows are
   hashed into a number of buckets which each has its own queue.  The
   number of buckets is configurable, and presently defaults to 1024 in
   the Linux implementation.  This is enough to avoid hash collisions on
   a moderate number of flows as seen for instance in a home gateway.
   Depending on the characteristics of the link, this can be tuned to
   trade off memory for a lower probability of hash collisions.  See
   Section 6 for a more in-depth discussion of this.

   By default, the flow hashing is performed on the 5-tuple of source
   and destination IP addresses and port numbers and IP protocol number.
   While the hashing can be customised to match on arbitrary packet
   bytes, care should be taken when doing so: Much of the benefit of the
   FQ-CoDel scheduler comes from this per-flow distinction.  However,
   the default hashing does have some limitations, as discussed in
   Section 6.



Hoeiland-Joergensen, etExpires September 19, 2016               [Page 6]

Internet-Draft                  fq-codel                      March 2016


   FQ-CoDel's DRR scheduler is byte-based, employing a deficit round-
   robin mechanism between queues.  This works by keeping track of the
   current number _byte credits_ of each queue.  This number is
   initialised to the configurable quantum; each time a queue gets a
   dequeue opportunity, it gets to dequeue packets, decreasing the
   number of credits by the packet size for each packet.  This continues
   until the value of _byte credits_ becomes zero or less, at which
   point it is increased by one quantum, and the dequeue opportunity
   ends.

   This means that if one queue contains packets of, for instance, size
   quantum/3, and another contains quantum-sized packets, the first
   queue will dequeue three packets each time it gets a turn, whereas
   the second only dequeues one.  This means that flows that send small
   packets are not penalised by the difference in packet sizes; rather,
   the DRR scheme approximates a (single-)byte-based fairness queueing
   scheme.  The size of the quantum determines the scheduling
   granularity, with the tradeoff from too small a quantum being
   scheduling overhead.  For small bandwidths, lowering the quantum from
   the default MTU size can be advantageous.

   Unlike plain DRR there are two sets of flows - a "new" list for flows
   that have not built a queue recently, and an "old" list for queues
   that build a backlog.  This distinction is an integral part of the
   FQ-CoDel scheduler and is described in more detail in Section 4.

4.  The FQ-CoDel scheduler

   To make its scheduling decisions, FQ-CoDel maintains two ordered
   lists of active queues, called "new" and "old" queues.  When a packet
   is added to a queue that is not currently active, that queue becomes
   active by being added to the list of new queues.  Later on, it is
   moved to the list of old queues, from which it is removed when it is
   no longer active.  This behaviour is the source of some subtlety in
   the packet scheduling at dequeue time, explained below.

4.1.  Enqueue

   The packet enqueue mechanism consists of three stages: classification
   into a queue, timestamping and bookkeeping, and optionally dropping a
   packet when the total number of enqueued packets goes over the
   maximum.

   When a packet is enqueued, it is first classified into the
   appropriate queue.  By default, this is done by hashing (using a
   Jenkins hash function [JENKINS]) on the 5-tuple of IP protocol, and
   source and destination IP addresses and port numbers (if they exist),
   and taking the hash value modulo the number of queues.  The hash is



Hoeiland-Joergensen, etExpires September 19, 2016               [Page 7]

Internet-Draft                  fq-codel                      March 2016


   salted by modulo addition of a random value selected at
   initialisation time, to prevent possible DoS attacks if the hash is
   predictable ahead of time (see Section 8).  The Linux kernel
   implements the Jenkins hash function by mixing three 32-bit values
   into a single 32-bit output value.  Inputs larger than 96 bits are
   reduced by additional mixing steps, 96 bits at a time.

   Once the packet has been successfully classified into a queue, it is
   handed over to the CoDel algorithm for timestamping.  It is then
   added to the tail of the selected queue, and the queue's byte count
   is updated by the packet size.  Then, if the queue is not currently
   active (i.e., if it is not in either the list of new or the list of
   old queues), it is added to the end of the list of new queues, and
   its number of credits is initiated to the configured quantum.
   Otherwise, the queue is left in its current queue list.

   Finally, the total number of enqueued packets is compared with the
   configured limit, and if it is _above_ this value (which can happen
   since a packet was just enqueued), a packet is dropped from the head
   of the queue with the largest current byte count.  Note that this in
   most cases means that the packet that gets dropped is different from
   the one that was just enqueued, and may even be from a different
   queue.

4.1.1.  Alternative classification schemes

   As mentioned previously, it is possible to modify the classification
   scheme to provide a different notion of a 'flow'.  The Linux
   implementation provides this option in the form of the "tc filter"
   command.  While this can add capabilities (for instance, matching on
   other possible parameters such as MAC address, diffserv code point
   values, firewall rules, flow specific markings, IPv6 flow label,
   etc.), care should be taken to preserve the notion of 'flow' as much
   of the benefit of the FQ-CoDel scheduler comes from keeping flows in
   separate queues.

   For protocols that do not contain a port number (such as ICMP), the
   Linux implementation simply sets the port numbers to zero and
   performs the hashing as usual.  In practice, this results in such
   protocols to each get their own queue (except in the case of hash
   collisions).  An implementation can perform other classifications for
   protocols that have their own notion of a flow, but SHOULD fall back
   to simply hashing on source and destination IP address and IP
   protocol number in the absence of other information.

   The default classification scheme can additionally be improved by
   performing decapsulation of tunnelled packets prior to hashing on the
   5-tuple in the encapsulated payload.  The Linux implementation does



Hoeiland-Joergensen, etExpires September 19, 2016               [Page 8]

Internet-Draft                  fq-codel                      March 2016


   this for common encapsulations known to the kernel, such as 6in4
   [RFC4213], IP-in-IP [RFC2003] and GRE (Generic Routing Encapsulation)
   [RFC2890].  This helps to distinguish between flows that share the
   same (outer) 5-tuple, but of course is limited to unencrypted tunnels
   (see Section 6.2).

4.2.  Dequeue

   Most of FQ-CoDel's work is done at packet dequeue time.  It consists
   of three parts: selecting a queue from which to dequeue a packet,
   actually dequeuing it (employing the CoDel algorithm in the process),
   and some final bookkeeping.

   For the first part, the scheduler first looks at the list of new
   queues; for the queue at the head of that list, if that queue has a
   negative number of credits (i.e., it has already dequeued at least a
   quantum of bytes), it is given an additional quantum of credits, the
   queue is put onto _the end of_ the list of old queues, and the
   routine selects the next queue and starts again.

   Otherwise, that queue is selected for dequeue.  If the list of new
   queues is empty, the scheduler proceeds down the list of old queues
   in the same fashion (checking the credits, and either selecting the
   queue for dequeuing, or adding credits and putting the queue back at
   the end of the list).

   After having selected a queue from which to dequeue a packet, the
   CoDel algorithm is invoked on that queue.  This applies the CoDel
   control law, which is the mechanism CoDel uses to determine when to
   drop packets (see [I-D.ietf-aqm-codel]).  As a result of this, one or
   more packets may be discarded from the head of the selected queue,
   before the packet that should be dequeued is returned (or nothing is
   returned if the queue is or becomes empty while being handled by the
   CoDel algorithm).

   Finally, if the CoDel algorithm does not return a packet, then the
   queue must be empty, and the scheduler does one of two things: if the
   queue selected for dequeue came from the list of new queues, it is
   moved to _the end of_ the list of old queues.  If instead it came
   from the list of old queues, that queue is removed from the list, to
   be added back (as a new queue) the next time a packet arrives that
   hashes to that queue.  Then (since no packet was available for
   dequeue), the whole dequeue process is restarted from the beginning.

   If, instead, the scheduler _did_ get a packet back from the CoDel
   algorithm, it subtracts the size of the packet from the byte credits
   for the selected queue and returns the packet as the result of the
   dequeue operation.



Hoeiland-Joergensen, etExpires September 19, 2016               [Page 9]

Internet-Draft                  fq-codel                      March 2016


   The step that moves an empty queue from the list of new queues to
   _the end of_ the list of old queues before it is removed is crucial
   to prevent starvation.  Otherwise the queue could reappear (the next
   time a packet arrives for it) before the list of old queues is
   visited; this can go on indefinitely even with a small number of
   active flows, if the flow providing packets to the queue in question
   transmits at just the right rate.  This is prevented by first moving
   the queue to _the end of_ the list of old queues, forcing a pass
   through that, and thus preventing starvation.  Moving it to the end
   of the list, rather than the front, is crucial for this to work.

   The resulting migration of queues between the different states is
   summarised in the following state diagram:

   +-----------------+                +------------------+
   |                 |     Empty      |                  |
   |     Empty       |<---------------+       Old        +----+
   |                 |                |                  |    |
   +-------+---------+                +------------------+    |
           |                             ^            ^       |Credits
           |Arrival                      |            |       |Exhausted
           v                             |            |       |
   +-----------------+                   |            |       |
   |                 |      Empty or     |            |       |
   |      New        +-------------------+            +-------+
   |                 | Credits exhausted
   +-----------------+

   Figure 1: Partial state diagram for queues between different states.
   Both the new and old queue states can additionally have arrival and
   dequeue events that do not change the state; these are omitted here.

5.  Implementation considerations

   This section contains implementation details for the FQ-CoDel
   algorithm.  This includes the data structures and parameters used in
   the Linux implementation, as well as discussion of some required
   features of the target platform and other considerations.

5.1.  Data structures

   The main data structure of FQ-CoDel is the array of queues, which is
   instantiated with the number of queues specified by the _flows_
   parameter at instantiation time.  Each queue consists simply of an
   ordered list of packets with FIFO semantics, two state variables
   tracking the queue credits and total number of bytes enqueued, and
   the set of CoDel state variables.  Other state variables to track




Hoeiland-Joergensen, etExpires September 19, 2016              [Page 10]

Internet-Draft                  fq-codel                      March 2016


   queue statistics can also be included: for instance, the Linux
   implementation keeps a count of dropped packets.

   In addition to the queue structures themselves, FQ-CoDel maintains
   two ordered lists containing references to the subset of queues that
   are currently active.  These are the list of 'new' queues and the
   list of 'old' queues, as explained in Section 4 above.

   In the Linux implementation, queue space is shared: there's a global
   limit on the number of packets the queues can hold, but not one per
   queue.

5.2.  Parameters

   The following are the user configuration parameters exposed by the
   Linux implementation of FQ-CoDel.

5.2.1.  Interval

   The _interval_ parameter has the same semantics as CoDel and is used
   to ensure that the minimum sojourn time of packets in a queue used as
   an estimator by the CoDel control algorithm is a relatively up-to-
   date value.  That is, CoDel only reacts to delay experienced in the
   last epoch of length interval.  It SHOULD be set to be on the order
   of the worst-case RTT through the bottleneck to give end-points
   sufficient time to react.

   The default interval value is 100 ms.

5.2.2.  Target

   The _target_ parameter has the same semantics as CoDel.  It is the
   acceptable minimum standing/persistent queue delay for each FQ-CoDel
   Queue.  This minimum delay is identified by tracking the local
   minimum queue delay that packets experience.

   The default target value is 5 ms, but this value should be tuned to
   be at least the transmission time of a single MTU-sized packet at the
   prevalent egress link speed (which for, e.g., 1Mbps and MTU 1500 is
   ~15ms), to prevent CoDel from being too aggressive at low bandwidths.
   It should otherwise be set to on the order of 5-10% of the configured
   interval.

5.2.3.  Packet limit

   Routers do not have infinite memory, so some packet limit MUST be
   enforced.




Hoeiland-Joergensen, etExpires September 19, 2016              [Page 11]

Internet-Draft                  fq-codel                      March 2016


   The _limit_ parameter is the hard limit on the real queue size,
   measured in number of packets.  This limit is a global limit on the
   number of packets in all queues; each individual queue does not have
   an upper limit.  When the limit is reached and a new packet arrives
   for enqueue, a packet is dropped from the head of the largest queue
   (measured in bytes) to make room for the new packet.

   In Linux, the default packet limit is 10240 packets, which is
   suitable for up to 10 Gigabit Ethernet speeds.  In practice, the hard
   limit is rarely, if ever, hit, as drops are performed by the CoDel
   algorithm long before the limit is hit.  For platforms that are
   severely memory constrained, a lower limit can be used.

5.2.4.  Quantum

   The _quantum_ parameter is the number of bytes each queue gets to
   dequeue on each round of the scheduling algorithm.  The default is
   set to 1514 bytes which corresponds to the Ethernet MTU plus the
   hardware header length of 14 bytes.

   In systems employing TCP Segmentation Offload (TSO), where a "packet"
   consists of an offloaded packet train, it can presently be as large
   as 64K bytes.  In systems using Generic Receive Offload (GRO), they
   can be up to 17 times the TCP max segment size (or 25K bytes).  These
   mega-packets severely impact FQ-CoDel's ability to schedule traffic,
   and hurt latency needlessly.  There is ongoing work in Linux to make
   smarter use of offload engines.

5.2.5.  Flows

   The _flows_ parameter sets the number of queues into which the
   incoming packets are classified.  Due to the stochastic nature of
   hashing, multiple flows may end up being hashed into the same slot.

   This parameter can be set only at initialisation time in the current
   implementation, since memory has to be allocated for the hash table.

   The default value is 1024 in the current Linux implementation.

5.2.6.  Explicit Congestion Notification (ECN)

   ECN is _enabled_ by default.  Rather than do anything special with
   misbehaved ECN flows, FQ-CoDel relies on the packet scheduling system
   to minimise their impact, thus the number of unresponsive packets in
   a flow being marked with ECN can grow to the overall packet limit,
   but will not otherwise affect the performance of the system.

   It can be disabled by specifying the _noecn_ parameter.



Hoeiland-Joergensen, etExpires September 19, 2016              [Page 12]

Internet-Draft                  fq-codel                      March 2016


5.2.7.  CE threshold

   This parameter enables Date Centre TCP (DCTCP)-like processing
   resulting in CE (Congestion Encountered) marking on ECN-Capable
   Transport (ECT) packets [RFC3168] starting at a lower sojourn delay
   setpoint than the default CoDel Target.  Details of DCTCP can be
   found in [I-D.ietf-tcpm-dctcp].

   The parameter, _ce_threshold_, is disabled by default and can be set
   to a number of microseconds to enable.

5.3.  Probability of hash collisions

   Since the Linux FQ-CoDel implementation by default uses 1024 hash
   buckets, the probability that (say) 100 flows will all hash to the
   same bucket is something like ten to the power of minus 300.  Thus,
   at least one of the flows will almost certainly hash to some other
   queue.

   Expanding on this, based on analytical equations for hash collision
   probabilities, for 100 flows, the probability of no collision is
   90.78%; the probability that no more than two of the 100 flows will
   be involved in any given collision = 99.57%; and the probability that
   no more than three of the 100 flows will be involved in any given
   collision = 99.99%.  These probabilities assume a hypothetical
   perfect hashing function, so in practice they may be a bit lower.  We
   have not found this difference to matter in practice.

   These probabilities can be improved upon by using set-associative
   hashing, a technique used in the Cake algorithm currently being
   developed as a further development upon the FQ-CoDel principles.  For
   a 4-way associative hash with the same number of total queues, the
   probability of no collisions for 100 flows is 99.93%, while for an
   8-way associative hash it is ~100%.

5.4.  Memory Overhead

   FQ-CoDel can be implemented with a low memory footprint (less than 64
   bytes per queue on 64 bit systems).  These are the data structures
   used in the Linux implementation:

   <CODE BEGINS>









Hoeiland-Joergensen, etExpires September 19, 2016              [Page 13]

Internet-Draft                  fq-codel                      March 2016


   struct codel_vars {
      u32             count;             /* number of dropped packets */
      u32             lastcount;     /* count entry to dropping state */
      bool            dropping;                /* currently dropping? */
      u16             rec_inv_sqrt;    /* reciprocal sqrt computation */
      codel_time_t    first_above_time;    /* when delay above target */
      codel_time_t    drop_next;                 /* next time to drop */
      codel_time_t    ldelay; /* sojourn time of last dequeued packet */
   };

   struct fq_codel_flow {
      struct sk_buff    *head;
      struct sk_buff    *tail;
      struct list_head  flowchain;
      int               credits;   /* current number of queue credits */
      u32               dropped; /* # of drops (or ECN marks) on flow */
      struct codel_vars cvars;
   };

   <CODE ENDS>

   The master table managing all queues looks like this:

   <CODE BEGINS>

   struct fq_codel_sched_data {
      struct tcf_proto *filter_list;  /* optional external classifier */
      struct fq_codel_flow *flows;    /* Flows table [flows_cnt] */
      u32             *backlogs;      /* backlog table [flows_cnt] */
      u32             flows_cnt;      /* number of flows */
      u32             perturbation;   /* hash perturbation */
      u32             quantum;        /* psched_mtu(qdisc_dev(sch)); */
      struct codel_params cparams;
      struct codel_stats cstats;
      u32             drop_overlimit;
      u32             new_flow_count;

      struct list_head new_flows;     /* list of new flows */
      struct list_head old_flows;     /* list of old flows */
   };

   <CODE ENDS>

5.5.  Per-Packet Timestamping

   The CoDel portion of the algorithm requires per-packet timestamps be
   stored along with the packet.  While this approach works well for
   software-based routers, it may be impossible to retrofit devices that



Hoeiland-Joergensen, etExpires September 19, 2016              [Page 14]

Internet-Draft                  fq-codel                      March 2016


   do most of their processing in silicon and lack space or mechanism
   for timestamping.

   Also, while perfect resolution is not needed, timestamp resolution
   finer than the CoDel target setting is necessary.  Furthermore,
   timestamping functions in the core OS need to be efficient as they
   are called at least once on each packet enqueue and dequeue.

5.6.  Limiting queueing in lower layers

   When deploying a queue management algorithm such as FQ-CoDel, it is
   important to ensure that the algorithm actually runs in the right
   place to control the queue.  In particular lower layers of the
   operating system networking stack can have queues of their own, as
   can device drivers and hardware.  Thus, it is desirable that the
   queue management algorithm runs as close to the hardware as possible.
   However, scheduling such complexity at interrupt time is difficult,
   so a small standing queue between the algorithm and the wire is often
   needed at higher transmit rates.

   In Linux, the mechanism to ensure these different needs are balanced
   is called "Byte Queue Limits" [BQL], which controls the device driver
   ring buffer (for physical line rates).  For cases where this
   functionality is not available, the queue can be controlled by means
   of a software rate limiter such as Hierarchical Token Bucket [HTB] or
   Hierarchical Fair-Service Curve [HFSC].  The Cake algorithm [CAKE]
   integrates a software rate limiter for this purpose.

   Other issues with queues at lower layers are described in [CODEL].

5.7.  Other forms of "Fair Queueing"

   Much of the scheduling portion of FQ-CoDel is derived from DRR and is
   substantially similar to DRR++. Versions based on Stochastic Fair
   Queueing [SFQ] have also been produced and tested in ns2.  Other
   forms of Fair Queueing, such as Weighted Fair Queueing [WFQ] or Quick
   Fair Queueing [QFQ], have not been thoroughly explored, but there's
   no a priori reason why the round-robin scheduling of FQ-CoDel
   couldn't be replaced with something else.

   For a comprehensive discussion of fairness queueing algorithms and
   their combination with AQM, see [I-D.ietf-aqm-fq-implementation].

5.8.  Differences between CoDel and FQ-CoDel behaviour

   CoDel can be applied to a single queue system as a straight AQM,
   where it converges towards an "ideal" drop rate (i.e., one that




Hoeiland-Joergensen, etExpires September 19, 2016              [Page 15]

Internet-Draft                  fq-codel                      March 2016


   minimises delay while keeping a high link utilisation), and then
   optimises around that control point.

   The scheduling of FQ-CoDel mixes packets of competing flows, which
   acts to pace bursty flows to better fill the pipe.  Additionally, a
   new flow gets substantial leeway over other flows until CoDel finds
   an ideal drop rate for it.  However, for a new flow that exceeds the
   configured quantum, more time passes before all of its data is
   delivered (as packets from it, too, are mixed across the other
   existing queue-building flows).  Thus, FQ-CoDel takes longer (as
   measured in time) to converge towards an ideal drop rate for a given
   new flow, but does so within fewer delivered _packets_ from that
   flow.

   Finally, the flow isolation FQ-CoDel provides means that the CoDel
   drop mechanism operates on the flows actually building queues, which
   results in packets being dropped more accurately from the largest
   flows than CoDel alone manages.  Additionally, flow isolation
   radically improves the transient behaviour of the network when
   traffic or link characteristics change (e.g., when new flows start up
   or the link bandwidth changes); while CoDel itself can take a while
   to respond, FQ-CoDel reacts almost immediately.

6.  Limitations of flow queueing

   While FQ-CoDel has been shown in many scenarios to offer significant
   performance gains compared to alternative queue management
   strategies, there are some scenarios where the scheduling algorithm
   in particular is not a good fit.  This section documents some of the
   known cases which either may require tweaking the default behaviour,
   or where alternatives to flow queueing should be considered.

6.1.  Fairness between things other than flows

   In some parts of the network, enforcing flow-level fairness may not
   be desirable, or some other form of fairness may be more important.
   An example of this can be an Internet Service Provider that may be
   more interested in ensuring fairness between customers than between
   flows.  Or a hosting or transit provider that wishes to ensure
   fairness between connecting Autonomous Systems or networks.  Another
   issue can be that the number of simultaneous flows experienced at a
   particular link can be too high for flow-based fairness queueing to
   be effective.

   Whatever the reason, in a scenario where fairness between flows is
   not desirable, reconfiguring FQ-CoDel to match on a different
   characteristic can be a way forward.  The implementation in Linux can
   leverage the packet matching mechanism of the _tc_ subsystem to use



Hoeiland-Joergensen, etExpires September 19, 2016              [Page 16]

Internet-Draft                  fq-codel                      March 2016


   any available packet field to partition packets into virtual queues,
   to for instance match on address or subnet source/destination pairs,
   application layer characteristics, etc.

   Furthermore, as commonly deployed today, FQ-CoDel is used with three
   or more tiers of service classification: priority, best effort and
   background, based on diffserv markings.  Some products do more
   detailed classification, including deep packet inspection and
   destination-specific filters to achieve their desired result.

6.2.  Flow bunching by opaque encapsulation

   Where possible, FQ-CoDel will attempt to decapsulate packets before
   matching on the header fields for the flow hashing.  However, for
   some encapsulation techniques, most notably encrypted VPNs, this is
   not possible.  If several flows are bunched into one such
   encapsulated tunnel, they will be seen as one flow by the FQ-CoDel
   algorithm.  This means that they will share a queue, and drop
   behaviour, and so flows inside the encapsulation will not benefit
   from the implicit prioritisation of FQ-CoDel, but will continue to
   benefit from the reduced overall queue length from the CoDel
   algorithm operating on the queue.  In addition, when such an
   encapsulated bunch competes against other flows, it will count as one
   flow, and not assigned a share of the bandwidth based on how many
   flows are inside the encapsulation.

   Depending on the application, this may or may not be desirable
   behaviour.  In cases where it is not, changing FQ-CoDel's matching to
   not be flow-based (as detailed in the previous subsection above) can
   be a mitigation.  Going forward, having some mechanism for opaque
   encapsulations to express to the outer layer which flow a packet
   belongs to, could be a way to mitigate this.  Naturally, care needs
   to be taken when designing such a mechanism to ensure no new privacy
   and security issues are raised by exposing information from inside
   the encapsulation to the outside world.  Keeping the extra
   information out-of-band and dropping it before it hits the network
   could be one way to achieve this.

6.3.  Low-priority congestion control algorithms

   In the presence of queue management schemes that limit latency under
   load, low-priority congestion control algorithms such as LEDBAT
   [RFC6817] (or, in general, algorithms that try to voluntarily use up
   less than their fair share of bandwidth) experiences little added
   latency when the link is congested.  Thus, they lack the signal to
   back off that added latency previously afforded them.  This effect is
   seen with FQ-CoDel as well as with any effective AQM [GONG2014].




Hoeiland-Joergensen, etExpires September 19, 2016              [Page 17]

Internet-Draft                  fq-codel                      March 2016


   As such, these delay-based algorithms tend to revert to loss-based
   congestion control, and will consume the fair share of bandwidth
   afforded to them by the FQ-CoDel scheduler.  However, low-priority
   congestion control mechanisms may be able to take steps to continue
   to be low priority, for instance by taking into account the vastly
   reduced level of delay afforded by an AQM, or by using a coupled
   approach to observing the behaviour of multiple flows.

7.  Deployment status and future work

   The FQ-CoDel algorithm as described in this document has been shipped
   as part of the Linux kernel since version 3.5, released on the 21st
   of July, 2012, with the ce_threshold being added in version 4.2.  The
   algorithm has seen widespread testing in a variety of contexts and is
   configured as the default queueing discipline in a number of mainline
   Linux distributions (as of this writing at least OpenWRT, Arch Linux
   and Fedora).  We believe it to be a safe default and encourage people
   running Linux to turn it on: It is a massive improvement over the
   previous default FIFO queue.

   Of course there is always room for improvement, and this document has
   listed some of the known limitations of the algorithm.  As such, we
   encourage further research into algorithm refinements and addressing
   of limitations.  One such effort is undertaken by the bufferbloat
   community in the form of the Cake queue management scheme [CAKE].  In
   addition to this we believe the following (non-exhaustive) list of
   issues to be worthy of further enquiry:

   o  Variations on the flow classification mechanism to fit different
      notions of flows.  For instance, an ISP might want to deploy per-
      subscriber scheduling, while in other cases several flows can
      share a 5-tuple, as exemplified by the RTCWEB QoS recommendations
      [I-D.ietf-tsvwg-rtcweb-qos].

   o  Interactions between flow queueing and delay-based congestion
      control algorithms and scavenger protocols.

   o  Other scheduling mechanisms to replace the DRR portion of the
      algorithm, e.g., QFQ or WFQ.

   o  Sensitivity of parameters, most notably the number of queues and
      the CoDel parameters.

8.  Security Considerations

   There are no specific security exposures associated with FQ-CoDel
   that are not also present in current FIFO systems.  On the contrary,
   some vulnerabilities of FIFO systems are reduced with FQ-CoDel (e.g.,



Hoeiland-Joergensen, etExpires September 19, 2016              [Page 18]

Internet-Draft                  fq-codel                      March 2016


   simple minded packet floods).  However, some care is needed in the
   implementation to ensure this is the case.  These are included in the
   description above, however we reiterate them here:

   o  To prevent packets in the new queues from starving old queues, it
      is important that when a queue on the list of new queues empties,
      it is moved to _the end of_ the list of old queues.  This is
      described at the end of Section 4.2.

   o  To prevent an attacker targeting a specific flow for a denial of
      service attack, the hash that maps packets to queues should not be
      predictable.  To achieve this, FQ-CoDel salts the hash, as
      described in the beginning of Section 4.1.  The size of the salt
      and the strength of the hash function is obviously a tradeoff
      between performance and security.  The Linux implementation uses a
      32 bit random value as the salt and a Jenkins hash function.  This
      makes it possible to achieve high throughput, and we consider it
      sufficient to ward off the most obvious attacks.

   o  Packet fragments without a layer 4 header can be hashed into
      different bins than the first fragment with the header intact.
      This can cause reordering and/or adversely affect the performance
      of the flow.  Keeping state to match the fragments to the
      beginning of the packet, or simply putting all packet fragments
      (including the first fragment of each fragmented packet) into the
      same queue, are two ways to alleviate this.

9.  IANA Considerations

   This document has no actions for IANA.

10.  Acknowledgements

   Our deepest thanks to Kathie Nichols, Van Jacobson, and all the
   members of the bufferbloat.net effort for all the help on developing
   and testing the algorithm.  In addition, our thanks to Anil Agarwal
   for his help with getting the hash collision probabilities in this
   document right.

11.  References

11.1.  Normative References

   [I-D.ietf-aqm-codel]
              Nichols, K., Jacobson, V., McGregor, A., and J. Iyengar,
              "Controlled Delay Active Queue Management", draft-ietf-
              aqm-codel-03 (work in progress), March 2016.




Hoeiland-Joergensen, etExpires September 19, 2016              [Page 19]

Internet-Draft                  fq-codel                      March 2016


   [I-D.ietf-aqm-fq-implementation]
              Baker, F. and R. Pan, "On Queuing, Marking, and Dropping",
              draft-ietf-aqm-fq-implementation-05 (work in progress),
              November 2015.

   [I-D.ietf-tcpm-dctcp]
              Bensley, S., Eggert, L., Thaler, D., Balasubramanian, P.,
              and G. Judd, "Datacenter TCP (DCTCP): TCP Congestion
              Control for Datacenters", draft-ietf-tcpm-dctcp-01 (work
              in progress), November 2015.

   [I-D.ietf-tsvwg-rtcweb-qos]
              Jones, P., Dhesikan, S., Jennings, C., and D. Druta, "DSCP
              and other packet markings for WebRTC QoS", draft-ietf-
              tsvwg-rtcweb-qos-14 (work in progress), March 2016.

   [RFC2003]  Perkins, C., "IP Encapsulation within IP", RFC 2003,
              DOI 10.17487/RFC2003, October 1996,
              <http://www.rfc-editor.org/info/rfc2003>.

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119,
              DOI 10.17487/RFC2119, March 1997,
              <http://www.rfc-editor.org/info/rfc2119>.

   [RFC2890]  Dommety, G., "Key and Sequence Number Extensions to GRE",
              RFC 2890, DOI 10.17487/RFC2890, September 2000,
              <http://www.rfc-editor.org/info/rfc2890>.

   [RFC3168]  Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
              of Explicit Congestion Notification (ECN) to IP",
              RFC 3168, DOI 10.17487/RFC3168, September 2001,
              <http://www.rfc-editor.org/info/rfc3168>.

   [RFC4213]  Nordmark, E. and R. Gilligan, "Basic Transition Mechanisms
              for IPv6 Hosts and Routers", RFC 4213,
              DOI 10.17487/RFC4213, October 2005,
              <http://www.rfc-editor.org/info/rfc4213>.

   [RFC6817]  Shalunov, S., Hazel, G., Iyengar, J., and M. Kuehlewind,
              "Low Extra Delay Background Transport (LEDBAT)", RFC 6817,
              DOI 10.17487/RFC6817, December 2012,
              <http://www.rfc-editor.org/info/rfc6817>.








Hoeiland-Joergensen, etExpires September 19, 2016              [Page 20]

Internet-Draft                  fq-codel                      March 2016


11.2.  Informative References

   [BLOAT]    Gettys, J., "Bufferbloat: Dark buffers in the Internet.",
              in IEEE Internet Comput. 15, 3,
              DOI http://dx.doi.org/10.1109/MIC.2011.56, 2011,
              <http://www.bufferbloat.net/attachments/27/
              IC-15-03-Backspace.pdf>.

   [BLOATWEB]
              "Bufferbloat web site", <https://www.bufferbloat.net>.

   [BQL]      Herbert, T., "Network Byte Queue Limits", August 2011,
              <https://lwn.net/Articles/454390/>.

   [CAKE]     "Cake comprehensive queue management system",
              <http://www.bufferbloat.net/projects/codel/wiki/Cake>.

   [CODEL]    Nichols, K. and V. Jacobson, "Controlling Queue Delay",
              July 2012, <http://queue.acm.org/detail.cfm?id=2209336>.

   [DRR]      Shreedhar, M. and G. Varghese, "Efficient Fair Queueing
              Using Deficit Round Robin", in IEEE/ACM Trans. Netw. 4, 3,
              June 1996,
              <http://users.ece.gatech.edu/~siva/ECE4607/presentations/
              DRR.pdf>.

   [DRRPP]    MacGregor, M. and W. Shi, "Deficits for Bursty Latency-
              critical Flows: DRR++", in Proceedings IEEE International
              Conference on Networks 2000 (ICON 2000), 2000,
              <http://ieeexplore.ieee.org/xpls/
              abs_all.jsp?arnumber=875803>.

   [GONG2014]
              Gong, Y., Rossi, D., Testa, C., Valenti, S., and D. Taht,
              "Fighting the bufferbloat: on the coexistence of AQM and
              low priority congestion control", in 2013 IEEE Conference
              on Computer Communications Workshops (INFOCOM WKSHPS),
              July 2014, <http://perso.telecom-
              paristech.fr/~drossi/paper/rossi14comnet-b.pdf>.

   [HFSC]     Stoica, I., Zhang, H., and T. Eugene, "Hierarchical fair-
              service curve", in Sigcomm 1997 proceedings, 1997,
              <http://conferences.sigcomm.org/sigcomm/1997/papers/
              p011.pdf>.

   [HTB]      "Hierarchical Token Bucket",
              <https://en.wikipedia.org/wiki/Token_bucket#Variations>.




Hoeiland-Joergensen, etExpires September 19, 2016              [Page 21]

Internet-Draft                  fq-codel                      March 2016


   [JENKINS]  Jenkins, B., "A Hash Function for Hash Table Lookup",
              1996, <http://www.burtleburtle.net/bob/hash/doobs.html>.

   [LINUXSRC]
              "Current FQ-CoDel Linux source code", <https://git.kernel.
              org/cgit/linux/kernel/git/torvalds/linux.git/tree/net/
              sched/sch_fq_codel.c>.

   [NS2]      "NS2 web site", <http://nsnam.sourceforge.net/wiki>.

   [NS3]      "NS3 web site", <https://www.nsnam.org/wiki>.

   [QFQ]      Checconi, F., Rizzo, L., and P. Valente, "QFQ: Efficient
              packet scheduling with tight guarantees", in IEEE/ACM
              Transactions on Networking (TON), 2013,
              <http://dl.acm.org/citation.cfm?id=2525552>.

   [SFQ]      McKenney, P., "Stochastic fairness queueing", published as
              technical report, 2002,
              <https://web.archive.org/web/20151003174154/
              http://www2.rdrop.com/~paulmck/scalability/paper/
              sfq.2002.06.04.pdf>.

   [SQF]      Bonald, T., Muscariello, L., and N. Ostallo, "On the
              impact of TCP and per-flow scheduling on Internet
              Performance", in IEEE/ACM transactions on Networking,
              April 2012, <http://perso.telecom-
              paristech.fr/~bonald/Publications_files/BMO2011.pdf>.

   [WFQ]      Demers, A., Keshav, S., and S. Shenker, "Analysis and
              simulation of a fair queueing algorithm", in SIGCOMM
              Comput. Commun. Rev., September 1989,
              <http://doi.acm.org/10.1145/75247.75248>.

Authors' Addresses

   Toke Hoeiland-Joergensen
   Karlstad University
   Dept. of Computer Science
   Karlstad  65188
   Sweden

   Email: toke.hoiland-jorgensen@kau.se








Hoeiland-Joergensen, etExpires September 19, 2016              [Page 22]

Internet-Draft                  fq-codel                      March 2016


   Paul McKenney
   IBM Linux Technology Center
   1385 NW Amberglen Parkway
   Hillsboro, OR  97006
   USA

   Email: paulmck@linux.vnet.ibm.com
   URI:   http://www2.rdrop.com/~paulmck/


   Dave Taht
   Teklibre
   2104 W First street
   Apt 2002
   FT Myers, FL  33901
   USA

   Email: dave.taht@gmail.com
   URI:   http://www.teklibre.com/


   Jim Gettys
   21 Oak Knoll Road
   Carlisle, MA  993
   USA

   Email: jg@freedesktop.org
   URI:   https://en.wikipedia.org/wiki/Jim_Gettys


   Eric Dumazet
   Google, Inc.
   1600 Amphitheater Pkwy
   Mountain View, CA  94043
   USA

   Email: edumazet@gmail.com














Hoeiland-Joergensen, etExpires September 19, 2016              [Page 23]