|
|
 |
|
Viewing report
|
|
|
|
Gigabit Ethernet Performance in Windows and Linux: The Impact of Client OS on Throughput
The Tolly Group, Dec 2002
In many companies, Gigabit Ethernet (GbE) is emerging as the de facto network standard to support high-performance applications. Windows 2000 remains the de facto standard for desktop PCs. Yet, as IT shops begin to migrate high-performance applications to GbE connections, they must do so with the knowledge that the host PCs can support the high-performance application and the high-speed connection in an efficient manner.
While Gigabit connections no doubt will speed bulk file transfers, they may not help at all with common tasks like Web or database access where the network presently is not the bottleneck. In the past, Microsoft operating systems presented bottlenecks as well. While Windows 2000 is vastly improved over Windows NT 4, benchmarking actual throughput is important.
This study quantifies the response time and throughput benefits of desktop Gigabit Ethernet running in three popular office operating system environments: Windows 2000 Professional, Windows XP Professional, and Red Hat Linux 8.0. Given the added requirements for error-free operation (and the potential impact of bit-errors on overall performance), the tests were conducted in certified Cat 6 cable environments.
Common desktop-to-server functions are benchmarked, including bulk file transfer, that are known to stress the network, as well as interactive transactions that simulate database access or Web browsing.
Previous testing has shown the need to benchmark Gigabit Ethernet adapters on PCs with the fastest available CPUs and 64-bit, 66-MHz PCI slots. As a result, engineers chose to use “high-end” PCs. Also, prior to running any application layer throughput test, the cabling should be tested for conformance to all pertinent standards. Category 6 conformance tests also were run prior to the application throughput testing.
The bulk of the testing performed used NetIQ Chariot 4.3 to provide measurable application layer traffic. Most tests for maximum throughput employed the “high performance throughput” script available with this version of Chariot. For tests that were concerned with response time of smaller file sizes, engineers used the “file send short” script.
|
|
|
|
