Network driver performance tuning, Optimizing server hardware and bios configuration, Windows server network driver – Dell Emulex Family of Adapters User Manual
Page 652: Table 3-10, Network driver performance

Emulex Drivers for Windows User Manual
P010077-01A Rev. A
3. Configuration
NIC Driver Configuration
652
Network Driver Performance Tuning
Optimizing Server Hardware and BIOS Configuration
Adapter performance can be improved by selecting a more efficient PCIe packet
payload size. If the system BIOS allows selection of a larger PCIe packet size, selecting
at least a 512-byte PCIe packet payload size provides the best efficiency for PCIe data
transfers. This may be an option in the server's system BIOS. The current value is
displayed in Device Manager on the Status property page for the adapter.
Most computers offer multiple distinct memory channels, which should be configured
for channel interleaving for optimal performance. Optimal interleaving is achieved by
using the exact same DIMM configuration for each memory channel. Check the
manufacturer's documentation and BIOS parameters for details about optimizing
memory bandwidth. Typically, all the DIMM slots must be populated to make use of all
the memory channels. As a general rule, more DIMMs provide better performance by
allowing a higher degree of memory-access interleaving to occur. However, some
servers decrease the memory speed if using more than 2 DIMMs per memory channel –
so it is important to understand the trade-off for a particular server platform.
Some servers may allow memory mirroring or memory sparing, where the total
memory is divided in half and each location is stored twice. This allows fault recovery
if one memory location detects an error, but it greatly reduces the perceived memory
bandwidth of the system.
Nearly any desktop or low-end server has enough memory bandwidth for the adapter
to support DMA at 20 Gb/s of data (10 Gb/s read, 10 Gb/s write). However, most of
the memory demands come from the processor accessing the data for either packet
copies in the non-offloaded networking stack or application. Increasing the clock speed
of the memory interface to the processor can be critical for achieving the best
networking performance. This interface may be the front side bus (FSB), Intel's QPI, or
AMD's HyperTransport.
Windows Server Network Driver
Table 3-10 describes ways to use various NIC driver properties and Microsoft Windows
properties to performance tune a system.
Table 3-10 Windows Server Performance Tuning Situations
Situation
Answer/Solution
There are a large number of short-lived TCP
connections such as web server or e-mail
server.
Enable RSS, increase number of RSS queues, and disable
TCP offload.
There are large data transfers such as to a
file server, web server with file downloads,
or an FTP server.
Use TCP connection offload.
There are large data transfers such as to a
backup server.
Enable jumbo packets, and use TCP offload.