Techniques for avoiding the high memory overheads found on many modern shared-memory multiprocessors are of increasing importance in the development of high-performance multiprocessor protocol implementations. One such technique is processor-cache affinity scheduling, which can significantly lower packet latency and substantially increase protocol processing throughput [30]. In this paper, we evaluate several aspects of the effectiveness of affinity-based scheduling in multiprocessor network protocol processing, under packet-level and connection-level parallelization approaches. Specifically, we evaluate the performance of the scheduling technique 1) when a large number of streams are concurrently supported, 2) when processing includes copying of uncached packet data, 3) as applied to send-side protocol processing, and 4) in the presence of stream burstiness and source locality, two well-known properties of network traffic. We find that affinity-based scheduling performs well under the...

Over the last years, interest for connecting small devices such as sensors to an existing network infrastructure such as the global Internet has steadily increased. Such devices often has very limited CPU and memory resources and may not be able to run an instance of the TCP/IP protocol suite.

Good network hardware performance is often squandered by overheads for accessing the network interface (NI) within a host. NIs that support user-level messaging avoid frequent operating system (OS) action yet unnecessary copying can still result in low performance. We explore improving application messaging performance by eliminating all unnecessary copies (minimal messaging). For minimal messaging, NIs must support address translation and must do so more richly than has been done in the past. NI address translation should flexibly support higher-level abstractions, map all user space, exploit translation locality, and degrade gracefully when locality is poor. We classify NI address translation implementations based on where the lookup and the miss handling are performed (CPU or NI). We present alternative designs and we consider how they interact with the OS. We provide simulation results that evaluate the alternative design points and we demonstrate feasibility with a real implement...

Abstract not found

OF THE DISSERTATION Path IDs: A Mechanism for Reducing Network Software Latency by Jonathan Simon Kay Doctor of Philosophy in Computer Science and Engineering University of California, San Diego, 1994 Professor Joseph Pasquale, Chair Network performance is important to an increasing number of computer applications. Similarly, network hardware performance is climbing to meet the demands of these new applications, with the advent of 100-megabit Ethernet, ATM, and fast processors to drive them. By contrast, network software latency is improving relatively slowly. It poses a barrier to the increasing use of networks for a wide variety of applications; as this dissertation shows, minimizing latency is more important to the minimization of processing time of the overall network workload than maximizing throughput. This dissertation introduces PathIDs, a means of reducing network latency. The latency problem is exacerbated both by the relative lack of research into latency op...

Security and privacy are growing concerns in the Internet community, due to the Internet's rapid growth and the desire to conduct business over it safely. This desire has led to the advent of several proposals for security standards, such as secure IP, secure HTTP, and the Secure Socket Layer. All of these standards propose using cryptographic protocols such as DES and RSA. Thus, the need to use encryption protocols is increasing. Shared-memory multiprocessors make attractive server platforms, for example as secure World-Wide Web servers. These machines are becoming more common, as shown by recent vendor introductions of platforms such as SGI's Challenge, Sun's SPARCCenter, and DEC's AlphaServer. The spread of these machines is due both to their relative ease of programming and their good price/performance. This paper is an experimental performance study that examines how encryption protocol performance can be improved by using parallelism. We show linear speedup for several different ...

We investigate the confrontation of load splitting and caching in high-performance parallel network forwarding systems. Our study demonstrates that hash-based load splitting schemes tend to significantly improve the temporal locality of the address stream submitted to a single routing engine (RE), which in turn greatly facilitates caching as a means of increasing system performance. We also show that the impact of locality on the efficiency of load balancing cannot be ignored: load balancing in a parallel forwarding system cannot be studied in isolation from the caching issues.

In this paper, we explore the benefits of processor cache affinity scheduling of parallelized network protocol processing. We find that affinity scheduling, which has not previously been shown to be of significant benefit to common applications, can provide large performance gain in the context of parallelized protocol processing. We conduct a set of multiprocessorexperiments designed to measure packet processing time in a UDP/IP/FDDI protocol stack in the x-kernel on an SGI Challenge XL multiprocessor. These measurements are then used to parameterize a combined simulation/analytic model of multiprocessor protocol processing. Our simulation results show that affinity scheduling can significantly reduce message delay associated with protocol processing, allowing a host to support a greater number of concurrent streams, to provide a higher maximum throughput to individual streams, and to decrease the end-to-end latency seen by an application. We find the reduction in end-to-end l...

We present a novel user-space and transaction-oriented protocol for use in high-performance distributed computing applications. The TRAP protocol is designed to support low-latency communication in multithreaded message-passing libraries. To demonstrate the viability of TRAP, we implemented and tested the fully operational protocol and conducted several experiments comparing its runtime performance with that of popular communication libraries. For fairness, the experiments did not exploit the TRAP protocol's multithreading capability at the application level. The results show that a TRAP-based communication library can, in various scenarios, outperform extant libraries based on in-kernel or user-level protocols. The results of these experiments also show that the TRAP protocol exhibits scalability characteristics, with respect to number of computing and communicating nodes, that is superior to regular in-kernel TCP/IP and other user-space protocol implementations. Keywords: Network Com...

Local area networks (LANs) are a popular means for connecting autonomous workstations together in a computational environment. LANs offer several advantages over traditional centralized computer systems, including better reliability, scalability, and cost. Certain limitations of LANs like their limited geographical span and the upper bound on the number of hosts that can be connected by a single LAN, can be overcome by interconnecting LANs together into an extended LAN or internetwork. Interconnection devices, such as bridges or routers, forward packets as necessary between the LANs, creating the illusion of a single large network. The performance of network interconnection devices is key to the success of extended LANs. Interconnection devices must selectively forward packets between LANs with minimal delay, so as to transparently extend the LAN. As networks increase in size and bandwidth, the need for efficient network interconnection devices increases. One of the goals of this the...