Scheduling data streams for low latency and high throughput on a Cray XC40 using Libfabric

Abstract

Summary Achieving efficient many‐to‐many communication on a given network topology is a challenging task when many data streams from different sources have to be scattered concurrently to many destinations with low variance in arrival times. In such scenarios, it is critical to saturate but not to congest the bisectional bandwidth of the network topology in order to achieve a good aggregate throughput. When there are many concurrent point‐to‐point connections, the communication pattern needs to be dynamically scheduled in a fine‐grained manner to avoid network congestion (links, switches), overload in the node's incoming links, and receive buffer overflow. Motivated by the use case of the Compressed Baryonic Matter experiment (CBM), we study the performance and variance of such communication patterns on a Cray XC40 with different routing schemes and scheduling approaches. We present a distributed Data Flow Scheduler (DFS) that reduces the variance of arrival times from all sources at least 30 times and increases the achieved aggregate bandwidth by up to 50%.

Lc‐Stream: An elastic scheduling strategy with latency constraints in geo‐distributed stream computing environments

“…Unfortunately, using deep learning models for prediction introduces additional overhead. In order to achieve fair network usage and reduce latency, the scheduling method proposed by Salem et al 38 runs on a group of senders and receivers, and channels the distribution of large‐capacity data streams with high throughput, so that streams from the same observation time are quickly gathered in the compute nodes.…”

Foreword to the Special Issue of the Cray User Group

“…Motivated by a use case from the Compressed Baryonic Matter (CBM) experiment, Salem et al address the problem of achieving efficient many‐to‐many communication for a given network topology when multiple data streams from different sources need to be scattered to multiple destinations concurrently and with a low variance in arrival times. They present a nonblocking, distributed data flow scheduler that is built on libfabric and is thus compatible with many modern interconnects.…”