Presto: Edge-based Load Balancing for Fast Datacenter Networks

By Keqiang He, Eric Rozner, Kanak B. Agarwal, Wes Felter, John B. Carter, Aditya Akella. Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, 2015

364 citations recorded by the source corpus.

Datacenter networks deal with a variety of workloads, ranging from latency-sensitive small flows to bandwidth-hungry large flows. Load balancing schemes based on flow hashing, e.g., ECMP, cause congestion when hash collisions occur and can perform poorly in asymmetric topologies. Recent proposals to load balance the network require centralized traffic engineering, multipath-aware transport, or expensive specialized hardware. We propose a mechanism that avoids these limitations by (i) pushing load-balancing functionality into the soft network edge (e.g., virtual switches) such that no changes are required in the transport layer, customer VMs, or networking hardware, and (ii) load balancing on fine-grained, near-uniform units of data (flowcells) that fit within end-host segment offload optimizations used to support fast networking speeds. We design and implement such a soft-edge load balancing scheme, called Presto, and evaluate it on a 10 Gbps physical testbed. We demonstrate the computational impact of packet reordering on receivers and propose a mechanism to handle reordering in the TCP receive offload functionality. Presto's performance closely tracks that of a single, non-blocking switch over many workloads and is adaptive to failures and topology asymmetry.

Subfields: Computer Networks and Communications, Information Systems

View Paper: https://www.semanticscholar.org/paper/12d3952369fd92db86d4a3fff0a6fa8de3509095

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Information Systems

Citations in the corpus

DRILL: Micro Load Balancing for Low-latency Data Center Networks (2017, methodology): “Hence, similar to prior work [35, 42], we optionally deploy a buffer in the host GRO layer to restore correct ordering.”
HULA: Scalable Load Balancing Using Programmable Data Planes (2016, methodology): “More recently, PRESTO [37] proposed splitting ﬂows into TSO (TCP Segment Ofﬂoad) segments of size 64KB and sending them on multiple paths.”
Shenango: Achieving High CPU Efficiency for Latency-sensitive Datacenter Workloads (2019, neutral): “A variety of efficient techniques have been proposed to resequence packets [29, 30, 33].”
Re-architecting datacenter networks and stacks for low latency and high performance (2017, background): “At the other end of the spectrum, researchers have been keen to tackle collisions due to per-flow ECMP with centralized scheduling in Hedera [2], via protocol changes in Packet Spraying [12] and Presto [21], or via network subflow-treatment in Conga [3] and switch-redesign of LocalFlow [35].”
Homa: a receiver-driven low-latency transport protocol using network priorities (2018, background): “Similar to prior work, we assume that bandwidth in the network core is sufficient to accommodate the offered load, and that the network supports efficient load-balancing [1, 10, 16], so that packets are distributed evenly across the available paths (we assume simple randomized perpacket spraying in ”