Chameleon

Abstract

This paper presents Chameleon, a cloud network providing both predictable latency and high utilization, typically two conflicting goals, especially in multi-tenant datacenters. Chameleon exploits routing flexibilities available in modern communication networks to dynamically adapt toward the demand, and uses network calculus principles along individual paths. More specifically, Chameleon employs source routing on the "queue-level topology", a network abstraction that accounts for the current states of the network queues and, hence, the different delays of different paths. Chameleon is based on a simple greedy algorithm and can be deployed at the edge; it does not require any modifications of network devices. We implement and evaluate Chameleon in simulations and a real testbed. Compared to state-of-the-art, we find that Chameleon can admit and embed significantly, i.e., up to 15 times more flows, improving network utilization while meeting strict latency guarantees.

Towards Understanding the Performance of Traffic Policing in Programmable Hardware Switches

“…Providing predictable latency in programmable networks requires the computation of performance bounds, in particular delay and throughput, in the network. Deterministic network calculus (DNC) [9], [10] is the main framework used by state-of-the-art solutions for predictable latency [1], [2], [5] to compute performance bounds. DNC is a system theory for communication networks based on the min-plus algebra.…”

“…2), a Ryu SDN controller [13] is deployed, running a custom application that configures the DUT with a certain number of flow rules, and police the traffic based on the parameters listed in Table II. We note that the parameter values are chosen based on the related state-of-the-art approaches [1], [5]. Configuring a DUT (i.e., modifying flow rules and meters) is done through OpenFlow 1.3 (green dashed line in Fig.…”

“…As it can be seen in Table III, Pica8 switches support significantly more meters compared to HP. Since state-of-theart approaches rely on a fine-grained flow control [6], the total number of flows on each switch in the network can easily grow up to several thousands [1] (e.g., over 2000 flow rules). Thus, having around 2000 meters may be insufficient for some use-cases.…”

“…Indeed, unexpected bursts of traffic entering the network consumes bandwidth, build up queues, and potentially leads to unexpected packet losses and hence degraded performance for other tenants. To avoid that, operators rely on policing and/or shaping of the traffic at the edge of the network, typically performed directly on the end hosts [1]- [4].…”

“…With the assumption that these inaccuracies can be perfectly modeled, we investigate how these would impact the performance of state-of-the-art solutions for providing predictable latency in programmable networks. Using the open-sourced code of the Chameleon system [1], [8] for cloud networks, we quantify how much these inaccuracies impact the number of tenants the system can accommodate while still providing its predictability guarantees. Astonishingly, we observe that Chameleon reduces the number of tenants it accepts by around 50% compared to a situation where switches are deemed perfect.…”