Towards Programmable Packet Scheduling

Abstract

Switches today provide a small set of scheduling algorithms. While we can tweak scheduling parameters, we cannot modify algorithmic logic, or add a completely new algorithm, after the switch has been designed. This paper presents a design for a programmable packet scheduler, which allows scheduling algorithms-potentially algorithms that are unknown today-to be programmed into a switch without requiring hardware redesign.Our design builds on the observation that scheduling algorithms make two decisions: in what order to schedule packets and when to schedule them. Further, in many scheduling algorithms these decisions can be made when packets are enqueued. We leverage this observation to build a programmable scheduler using a single abstraction: the push-in first-out queue (PIFO), a priority queue that maintains the scheduling order and time for such algorithms.We show that a programmable scheduler using PIFOs lets us program a wide variety of scheduling algorithms. We present a detailed hardware design for this scheduler for a 64-port 10 Gbit/s shared-memory switch with <4% chip area overhead on a 16-nm standard-cell library. Our design lets us program many sophisticated algorithms, such as a 5-level hierarchical scheduler with programmable scheduling algorithms at each level.

NUMFabric

“…Swift switches do the same, except that the flow's weight is allowed to change on a packet-by-packet basis. We leave the details of the algorithm and its practical realization using recently proposed hardware mechanisms [59] to §5.…”

“…While today's switches do not support priority queues, recent work [59,60] has shown that it is feasible to implement programmable priority queues in emerging programmable switching chips [11]. This design allows packets to be inserted into the queue based on a programmable rank value that is computed before the packet is enqueued.…”

“…NUMFabric's packet scheduling differs from standard WFQ in that the flow's weights are set dynamically at the hosts and carried in packet headers. This design, which we call the Swift transport ( §4.1), exploits the fact that datacenter switches can be designed to support more sophisticated packet scheduling than simple FIFO queues [59]. Swift guarantees that the network is fully utilized while keeping flows isolated using WFQ.…”

Programmable Packet Scheduling at Line Rate

“…Moreover, for most algorithms used in practice, these two decisions can be determined definitively when a packet is enqueued into the packet buffer [38].…”

“…Furthermore, for many scheduling algorithms, these two decisions can be made when a packet is enqueued. This observation suggests a natural hardware primitive for packet scheduling: a push-in first-out queue (PIFO) [19,38]. A PIFO is a priority queue that allows elements to be pushed into an arbitrary position based on an element's rank (the scheduling order or time), 1 but always dequeues elements from the head.…”