High Performance Commodity Network Interfaces
نویسندگان
چکیده
As commodity workstations become faster with increases in processor technology, the need for higher performance network interfaces becomes more pressing. While current commodity network interface cards provide ample bandwidth for bulk data transfers, the latency imposed on small messages can be large. However, to attempt to harness the potential compute power of a cluster of workstations at commodity prices, attention must be paid to the way in which the nodes are interconnected. The high latency incurred by standard protocol stacks on commodity hardware makes clusters of workstations appropriate only for problems that exhibit extremely high computation to communication ratios. Projects such as GAMMA from the University of Genoa 8] have shown that by reducing the operating system overhead, the latency associated with communications can be signiicantly reduced, even on commodity hardware. Other projects, such as Myricom's Myrinet 6], and DEC/Compaq's Memory Channel 14], show that a different network interface architecture can also yield significant performance gains. By taking advantage of modern, high performance, highly integrated microcontrollers, future network card designs could realize even higher performance through lower operating system overhead and more capable network interfaces. 1 Introduction In recent years, the interest in achieving high performance computing on networks of workstations 2] has grown, as the power of commodity processors has increased. Even though processors continue to increase in performance as predicted by Moore's Law, the price of machines has dropped, making the cost-eeectiveness of cluster computing, as measured in compute performance per dollar spent, increase even faster. However, to exploit the high performance potential of modern processors, it is necessary to nd a way to interconnect the processing elements so that they may communicate ef-ciently to work together on large problems. In a custom designed highly parallel computer, a signiicant portion of time and resources goes into the design of the interconnec
منابع مشابه
A Survey of Trends in Fast Packet Processing
Network interfaces of 10 Gbit/s are used more and more. Despite their high performance, packet processing rates are not correspondingly higher. This happens due to the architecture of the network stacks and the design of device drivers. There are restrictive factors and overheads which need to be overcome, so as to enable fast packet processing with commodity hardware. In this paper, techniques...
متن کاملIncreasing Performance in Commodity IP
Clusters built from commodity hardware and software have several advantages over more traditional supercomputers. Commodity clusters are cheap and ubiquitous, and so they are easier to design, program and maintain. However, even as high-speed networks reach 10Gb/s speeds and modern computer architectures reach past 1Ghz, new commodity clusters are not able to harness this power. One of the bott...
متن کاملDesign, Implementation, and Performance of MPI on Portals 3.0
The emergence of cluster computing as a viable platform for high performance computing has been realized due to significant performance increases in commodity computing and networking hardware. In particular, relatively inexpensive programmable network interface cards (NICs), such as Myrinet (Boden et al., 1995), that are capable of delivering gigabit-per-second speeds, have allowed for much re...
متن کاملPerformance analysis of multi-radio AODV in hybrid wireless mesh networks
Wireless Mesh Networks (WMNs), based on commodity hardware, present a promising technology for a wide range of applications due to their self-configuring and self-healing capabilities, as well as their low equipment and deployment costs. One of the key challenges that WMN technology faces is the limited capacity and scalability due to co-channel interference, which is typical for multi-hop wire...
متن کاملüberSpark: Enforcing Verifiable Object Abstractions for Automated Compositional Security Analysis of a Hypervisor
We present überSpark (üSpark), an innovative architecture for compositional verification of security properties of extensible hypervisors written in C and Assembly. üSpark comprises two key ideas: (i) endowing low-level system software with abstractions found in higher-level languages (e.g., objects, interfaces, function-call semantics for implementations of interfaces, access control on interf...
متن کامل