High-performance implementation of in-network traffic pacing

Abstract

Optical packet switching networks promise to pro vide high-speed data communication and serve as the foundation of the future Internet. A key technological problem is the very small size of packet buffers that can be implemented in the optical domain. Existing protocols, for example the transmission control protocol, do not perform well in such small-buffer networks.To address this problem, we have proposed techniques for actively pacing traffic to ensure that traffic bursts are reduced or eliminated and thus do not cause packet losses in routers with small buffers. In this paper, we present the design and prototype of a hardware implementation of a packet pacing system based on the NetFPGA system. Our results show that traffic pacing can be implemented with few hardware resources and without reducing system throughput. Therefore, we believe traffic pacing can be deployed widely to improve the operation of current and future networks.Index Terms-small-buffer network, traffic pacing, traffic burstiness, field-programmable gate array, prototype I. IN T RODUC TIONHigh-bandwidth data communication is one of the key foundations for the future Internet. Society's expectations to perform many daily activities over the Internet (e.g., entertain ment, person communication, e-government, health care, etc.) has increased the demand for raw bandwidth in the network. One technology that promises to meet these demands for bandwidth is optical networking. Optical fiber has long been used for long-haul, point-to-point data transmissions. However, the need to convert from the optical domain back to the electronic domain for switching has limited the throughput that can be achieved optical-electronic-optical (OEO) networks.To address these performance limitations, all-optical net works (AON) have been developed. Optical circuit switching (OCS) can be used to provide end-to-end light paths. While OCS has been successfully deployed for some use case scenar ios, the slow timescales at which connections can be set up and tom down does not match the timescales of highly dynamic end-to-end Internet connections. As alternative, optical packet switching (OPS) [1] and optical burst switching (OBS) [2], [3] has been proposed. In these networks, packets or short bursts of packets are switched independently in the optical domain.While OPS and OBS are promising approaches to optical networks that match the needs of current Internet protocols and dynamics, there are important technological challenges that need to be addressed. Apart from the difficulties of building a switch that operates in the optical domain, there are also challenges imposed by the basic operation of the Internet. Packet switching in the Internet is based on statistical multiplexing and traffic is forwarded opportunistically without prior resource reservations. As a result, traffic may compete for link bandwidth on the output port of a switch. To avoid packet losses, packet buffers are used to queue packets and thus absorb short periods of overload. In electronic networks an...

The Case for Moving Congestion Control Out of the Datapath

“…Given the advent of programmable NICs [14,39,50], we anticipate that supporting measurement fold functions and control programs should be feasible. Rate-based congestion control algorithms can be implemented using packet pacing in hardware [25,35,43]. A key impediment to supporting CCP is maintaining per-flow congestion state on the NIC.…”

High-performance implementation of in-network traffic pacing for small-buffer networks

“…In our work, we implement a pacer system in programmable hardware. Preliminary results of this work have been published in [20].…”

On Hardware Programmable Network Dynamics With a Chemistry-Inspired Abstraction

“…[11], [29], [30], [31]), except for a narrow segment of non-novices in hardware programming (e.g. [1], [32]).…”