Accelerating Communication-Intensive Parallel Workloads Using Commodity Optical Switches and a Software-Configurable Control Stack
نویسندگان
چکیده
In response to the need for faster and fatter networks for large-scale HPC cluster systems, hybrid optical/electrical networks have been proposed as an affordable and high-capacity solution. Still, there is no prior work evaluating the performance of HPC workloads over such types of networks. To fill this gap, this work presents a hybrid network architecture comprising commodity-only equipment, shows its price competitiveness against fat-tree alternatives and presents a prototype implementation. We evaluated several HPC workloads over our prototype, showing that our hybrid optical/electrical network manages to significantly accelerate tested workloads, without incurring any extra cost compared to an all-electronic fat-tree network.
منابع مشابه
VAMOS: Virtualization Aware Middleware
Machine virtualization is undoubtedly useful, but does not come cheap. The performance cost of virtualization, for I/O intensive workloads in particular, can be heavy. Common approaches to solving the I/O virtualization overhead focus on the I/O stack, thereby missing optimization opportunities in the overall stack. We propose VAMOS, a novel software architecture for middleware, which runs midd...
متن کاملEmploying transport layer multi-railing in cluster networks
Building clusters from commodity off-the-shelf parts is a well-established technique for building inexpensivemediumto large-size computing clusters.Many commoditymid-rangemotherboards comewith multiple Gigabit Ethernet interfaces, and the low cost per port for Gigabit Ethernet makes switches inexpensive as well. Our objective in this work is to take advantage of multiple inexpensive Gigabit net...
متن کاملSoftware-Controlled Next Generation Optical Circuit Switching for HPC and Cloud Computing Datacenters
In this paper, we consider the performance of optical circuit switching (OCS) systems designed for data center networks by using network-level simulation. Recent proposals have used OCS in data center networks but the relatively slow switching times of OCS-MEMS switches (10–100 ms) and the latencies of control planes in these approaches have limited their use to the largest data center networks...
متن کاملScalable Inter-Cluster Communication Systems for Clustered Multiprocessors
As workstation clusters move away from uniprocessors in favor of multiprocessors to support the increasing computational needs of distributed applications, greater demands are placed on the communication interfaces that couple individual workstations. This paper investigates scalable, e cient, and reliable communication systems for multiprocessor clusters that use commodity local area networks ...
متن کاملActive I/O Switches in System Area Networks
We present an active switch architecture to improve the performance of systems connected via system area networks. Our programmable active switches not only flexibly route packets between any combination of hosts and I/O devices, but also have the capability of running application-level code, forming a parallel processor in the SAN subsystem. By replacing existing SAN-based switches with a new ...
متن کامل