Home | english  | Impressum | Datenschutz | Sitemap | KIT

Flexible IRT Scheduling for Software-Defined Networking

Flexible IRT Scheduling for Software-Defined Networking
Forschungsthema:SDN, Industrie 4.0, TSN, IRT-Scheduling
Typ:Masterarbeit
Datum:ab sofort
Betreuer:

Tim Gerhard

Description

IEEE Time-Sensitive Networking offers the ability to transmit best-effort traffic, QoS traffic, and isochronous real-time (IRT) traffic over the same network. While best-effort or QoS are easy to route flexibly, the scheduling of IRT traffic is an NP-hard problem. Moreover, existing schedules cannot easily be completely replaced during runtime without violating real-time guarantees.

These restrictions make it hard to integrate new real-time flows into an existing network, either due to connecting new devices, or due to changed communication relationships between existing devices. The latter case may be solved by providing enough bandwidth to serve all possible real-time streams, but overprovisioning is costly and should not be the only solution to this problem.

In our case, we need to group streams into processes, so that streams can be reserved in an all-or-nothing manner where applicable. A schedule shall be calculated for an initial set of enabled processes, realizing as many of them as possible in the given network. From this initial schedule, processes may be enabled or disabled, and the network should alter the current schedule to realize the updated requirement without recalculating everything.

From this point, it has to be evaluated whether, and if yes how likely, the enabling and disabling of processes can lead to ineffective schedules, i.e. by having an active data stream blocking certain time slots that would be needed by other streams.

Assignment

In the first part of this thesis, the scheduling problem shall be solved, e.g. by formulating an ILP problem that models the grouping of streams into processes.

In the next step, enabling and disabling of individual processes shall be solved. This can be done by either scheduling any combination of proccesses in advance, or dynamically add or remove the respective streams, iff possible. Both solutions should be implemented and compared to each other.

For the evaluation of this approach, an \emph{efficiency} metric has to be developed and applied to selected test scenarios.

Alternative: Only implement the adding/removing to/from an existing schedule, and find ways to alter an existing schedule without causing jitter or packet loss. In this case, use a metric to decide when to transition between schedules.

Prerequisites

  • Integer Linear Programming (ILP)
  • Programming Expertise, Understanding of computational complexity
  • Any programming language