Active Queue Management and Packet Scheduling
charter-ietf-aqm-00-01

Internet routers, lower-layer switches, end-host operating
systems, device drivers, and many types of additional
middleboxes include memory buffers in which they implement
queues to hold packets that require processing or otherwise
need to wait for forwarding to the next hop.

The queues are intended to absorb bursts of traffic that may
naturally occur, and avoid unneccessary losses. However, queues
also cause latency and jitter in the eventual arrival times of
packets. This can create issues and complications for interactive
applications.

Extremely large unmanaged buffers have been noticed in some
software and equipment. When these buffers fill, interactive
applications and other traffic can be severely impacted or
completely broken, due to high and potentially oscillating delays.

The Active Queue Management and Packet Scheduling working group
(AQM) works on algorithms for managing queues in order to:

(1) minimize standing queues, helping to reduce delay for
interactive applications

(2) help flow sources control their sending rates without
unnecessary losses, e.g. through ECN

(3) consider the merits of various techniques to protect flows
from negative impacts of other more aggressive or misbehaving
flows

(4) help avoid global synchronization of flows sharing a
bottleneck

The AQM working group will produce Informational, Best Current
Practices, and Standards Track Applicability Statement documents
that cover the design, use, and configuration of algorithms for
managing queues in Internet devices and software. The scope
includes both how to best configure existing equipment and
software, as well as recommendations on designing new equipment
and software.

The AQM working group will also publish algorithm specifications
that are found to be broadly applicable and beneficial. Evaluating
these algorithms shall be done in coordination with the Internet
Congestion Control Research Ground (ICCRG), and related IETF Working
Groups, such as the RTP Media Congestion Avoidance Techniques Working
Group (RMCAT), in order to select and assess the relevant criteria,
scenarios, and metrics.

The working group will also explore the merits of whether to
isolate flows, and mechanisms for performing this function. Note
that isolation and potentially policing of flows implies some policy
beyond what is required to simply minimize queues. This topic
requires significant attention in the working group.

AQM algorithms do not have to be implemented universally in order
to be effective. Specifications will aid in producing proper
implementations that avoid potential ambiguities and corner cases.
"Interoperability" of algorithms and implementations of them is
not the reason for creating these specifications; correctness is
the primary motivation.

The working group will not make changes to existing IETF protocols,
but the working group may use ECN, Diffserv, and other mechanisms
maintained by the TSVWG working group. Since the implementation of
these mechanisms is likely to be entwined with AQM algorithms, there
is expected to be close coordination between the TSVWG and AQM groups.

Many AQM algorithms have been proposed in academic literature, but
a smaller number are widely implemented and deployed. The goal of the
working group is to produce recommendations that will actually be used,
and algorithms that will actually be implemented, deployed in equipment,
and enabled. Towards these ends, the group actively encourages
participation from operators and implementers. Furthermore, the group
will jointly work with the Routing and Internet Area in order to
involve vendors of networking equipment in the development of the
AQM mechanisms.

Wider research and evaluation of AQM mechanisms shall be
coordinated with the IRTF/ICCRG, and significant participation in this
WG from the academic and research community is highly desirable, when
it is directly relevant to implementation and deployment.

Combined Queue Management / Packet Scheduling algorithms are in-scope,
provided their benefit have been evaluated against the established
requirements for an AQM algorithm. It is expected that some classes of
algorithms will focus on software implementations, while others on
existing or new hardware deployments, and algorithms may be specific
to distinct scenarios.