Steve Pope

Distributed computing with the CLAN network

CLAN (collapsed LAN) is a high performance user-level network targeted at the server room. It presents a simple low-level interface to applications: connection-oriented non-coherent shared memory for data transfer, and Tripwire, a user-level programmable content addressable memory (CAM) for synchronisation. This simple interface is implemented using only hardware state machines on the network interface controller (NIC), yet is flexible enough to support many different applications and communications paradigms. We show how CLAN is used to support a number of standard transport protocols and middleware: MPI, VIA, TCP/IP and CORBA. In each case we demonstrate performance that approaches,the underlying network. For TCP/IP we present our initial results using an in-kernel stack, and describe the architecture of our prototype Gigabit Ethernet/CLAN bridge, which demultiplexes Ethernet frames directly to user-level TCP/IP stacks via the CLAN network. For VIA we present a software implementation with better latency than a commercial VIA NIC implemented on ASIC technology.

The design and implementation of a high-speed user-space transport protocol

Audio and video are fast becoming an integral part of new computing environments. These media have transport requirements which differ from the normal bursty computer traffic. There is therefore a need to explore transport protocols that can provide different qualities of service. User-space implementations of such protocols are particularly interesting because they can be easily tested and refined. This paper discusses the design and implementation of a high-speed user-space protocol called A1. Its preliminary performance in an ATM environment is presented and compared with an efficient kernel implementation of TCP/IP.

FPGA augmented ASICs: The time has come

This article consists of a collection of slides from the authors conference presentation on field programmable gate arrays (FPGA) acceleration. Some of the specific topics discussed include: an overview of the Ivy Bridge architecture and supported applications; power scaling and management facilities; core product features; power efficiency; voltage control and optimization technqiues; power sharing capabilities; and system architecture.

TRIPWIRE: A SYNCHRONISATION PRIMITIVE FOR VIRTUAL MEMORY MAPPED COMMUNICATION

Existing user-level network interfaces deliver high bandwidth, low latency performance to applications, but are typically unable to support diverse styles of communication and are unsuitable for use in multiprogrammed environments. Often this is because the network abstraction is presented at too high a level, and support for synchronisation is inflexible. In this paper we present a new primitive for in-band synchronisation: the Tripwire. Tripwires provide a flexible, efficient and scalable means for synchronisation that is orthogonal to data transfer. We describe the implementation of a non-coherent distributed shared memory network interface, with Tripwires for synchronisation. This interface provides a low-level communications model with gigabit class bandwidth and very low overhead and latency. We show how it supports a variety of communication styles, including remote procedure call, message passing and streaming.