(Toward 1 Billion IPv6 Searches per Second
: Page 1 of 1 )
The ability of an Internet Protocol network to provide high-performance search results is being stressed by several factors, including:
The massive deployment of wireless devices, each with its own IP address, which require ever-larger lookup tables;
The demand for triple-play services--voice, video and data--at competitive prices;
The migration of IPv4 to IPv6, which involves moving from 32-bit search addresses to 128-bit; and
Other advanced infrastructure requirements, such as multilayer Ethernet protocols, improved security and access control.
A dramatic increase in network search capability is crucial to the high-performance, cost-efficient delivery of these services. Dedicated search accelerators accomplish that increase by offloading search-processing tasks from network processors and other devices.
This article discusses how effectively various search accelerator architectures deliver the requisite results, and how to achieve up to 1 billion IPv6 searches per second.
Increasing the Search Rate
A search accelerator's search rate is a product of several variables. One key factor is how the device drives the search clock. Single-data-rate (SDR) devices use a traditional 2x clocking scheme, which transfers data only on the rising edges of the 2x clock signal. This scheme requires at least two 2x clock cycles to transfer a complete search key, so the search rate is half the bus clock rate. For instance, to achieve a performance of 125 Msamples/second, the device must operate at 250 MHz.
One architectural advancement on this approach is the double-data-rate interface, which transfers data on both the rising and falling edges of the clock signal. DDR devices enable the clock input to drive the same search rate as the ternary content-addressable memory (TCAM) core. For example, a 250-Msample/s search rate can be achieved with a 250-MHz clock, instead of a 500-MHz clock.
The lower clock rate not only reduces power consumption but also allows the use of lower-cost components.
2008 International Mathematica Conference Dr. Dobb's interviews Wolfram Research's Theo Gray, co-founder and Director of User Interfaces, and Roger Germundsson, Director of Research and Development, about the upcoming 2008 International
Mathematica Conference.
How Do You Do Nightly Builds and Tests when there is No Overnight? Software Production in a Geographically Distributed Environment
Attend this Webcast and find out how to overcome common build-test-deploy challenges that affect all members of a distributed team, including:
<ul>
<li> Communication difficulties, because of time-zone and cultural differences</li>
<li> Workflow challenges, like lack of documented procedures and build and test handoff problems</li>
<li> Slow build and test cycles, broken builds, and other factors that hamper distributed team productivity</li>
</ul>
Thursday, September 25, 2005 " 11am PT / 2pm ET
</p>
In this volume of Best of BYTE, we explore the emergence of some heuristic algorithms. Although we have only scratched the surface of this intriguing subject, we hope we've suggested the potential of the synthesis of heuristics and algorithms.
Understand C/C++ code in less time. A new team member ? Inherited legacy code ? Get up to speed faster with Crystal Flow for C/C++. Code-formatting improves readability. Flowcharts are integrated with code browser. Export flowcharts to Visio.