Skip to main content

Computing-Aware Networking

Document Proposed charter Computing-Aware Networking WG (can)
Title Computing-Aware Networking
Last updated 2023-03-03
State Replaced
WG State Replaced
IESG Responsible AD John Scudder
Charter edit AD John Scudder
Send notices to (None)

Computing-Aware Traffic Steering (cats)

Many service architectures create multiple service instances. These
instances are often geographically distributed to multiple sites, and
a single site may support multiple instances of a service. The services
are provided on computing platforms and are generically
referred to as "compute services". The CATS (Computing-Aware Traffic Steering)
working group (WG) is chartered to consider the problem of how the
network edge can steer traffic between clients of a service and sites
offering the service.

Since, for some services (for example, the evolution of networked AR/VR,
and deployment of autonomous and networked vehicles), the performance
experienced by clients will depend on both network metrics such as
bandwidth and latency, and compute metrics such as processing, storage
capabilities, and capacity, there is a need for a solution that can
optimize how a network edge node steers traffic based on these metrics,
as appropriate to the service.

Although the specific optimization function will likely differ between
services, implementations, and deployments, there is a need for a
general framework for the distribution of compute and network metrics
and transport of traffic from network edge to service instance. It also
is likely that some set of common metrics can be identified. The CATS WG
will concern itself with these issues.

The IETF is working on exposing network conditions to endpoints
(notably ALTO) and load balancing/service selection at layers 4 and 7
(for example, related to the selection of SIP servers). Specific
characteristics that may distinguish CATS from other work include the
desire to integrate both network and compute conditions in the
optimization function that informs the steering applied by the network
edge nodes, and the desire to operate that function on nodes within the
service provider's network, logically separated from service operation.
Exposure of network and compute conditions to applications is not in the
scope of CATS. Because of their experience and prior work in collecting
and exposing network conditions for use in selecting paths and servers,
the CATS WG will seek advice and expertise from the ART and TSV areas.

The assumed model for the CATS WG is an overlay network, where a network
edge node makes a decision based on the metrics of interest, and then
steers the traffic to a node that serves a service instance, for example
using a tunnel. The CATS WG will focus on single domain models.
Architectures that require the underlay network to be service-aware
are out of scope.

The CATS WG will analyze the problem in further detail and produce an
architecture for a solution. Ideally, that architecture will be one that
can be instantiated using existing technologies.

The CATS WG is chartered to work on the following items:

o Groundwork may be documented via a set of informational Internet-
  Drafts, not necessarily for publication as RFCs:

  * Problem statement for the need to consider both network and
    computing resource status.

  * Use cases for steering traffic from applications that have critical
    SLAs that would benefit from the integrated consideration of network
    and computing resource status.

  * Requirements for commonly agreed computing metrics and their
    distribution across the overlay network, as well as the appropriate
    frequency and scope of distribution.

o Overall CATS framework & architecture:

  * This work encompasses the various building blocks and their
    interactions, realizing a CATS control and data plane that addresses
    the identified problems and requirements in the groundwork,
    including methods for distributing necessary information to utilize
    the identified metrics in CATS use cases. This will also cover OAM,
    scalability, and security aspects.

o Additional groundwork to include:

  * Analyze the suitability and usefulness of computing and networking
    metrics for traffic steering decisions in CATS with a CATS metrics
    ontology as a possible outcome.

  * Analyze methods for distributing the necessary information to
    utilize the identified metrics in CATS use cases.

o Applicability of existing tools and mechanisms:

  * Analysis of implementing the CATS control and data plane using
    existing mechanisms, including identifying the limitations of
    existing tools in fulfilling requirements.

  * Study potential new approaches for the CATS control and data plane
    solution that can fill the identified gaps in existing tools and

  * Study FCAPS (fault, configuration, accounting, performance,
    security) requirements, mechanisms, and suitability of existing
    messaging protocols (NETCONF) and data models (YANG).