Windows Server 2008 includes a new implementation (a complete redesign) of the original TCP/IP protocol stack called the Next Generation TCP/IP stack. This new framework is a total rewrite of TCP/IP functionality for both IPv4 and IPv6. It’s designed to better meet connectivity and performance needs in various networking environments using various networking technologies.
For the benefit of those stuck in a cave in Patagonia since the early 1980s, TCP/IP is the de facto standard network protocol stack for most server and workstation computers you’ll encounter, but it’s by no means the only one. It expands to Transmission Control Protocol/Internet Protocol and serves as the foundation for network traffic shuttled across the Internet. It’s become a nearly universal means for networked communications of all kinds.
The core network stack framework is improved and enhanced to increase existing functionality, complement it with supplementary performance enhancing functionality, and further expand that framework through additional features and components. The following are material that’s both directly and indirectly related to advances in the Next Generation TCP/IP network protocol stack in Windows Server 2008.
Receive window auto-tuning
In TCP, a receive window size defines the amount of data that a TCP receiver permits a TCP sender to push onto the network before requiring the sender to wait for acknowledgement of its receipt. Correctly determining the maximum receive window size for a connection is now automatically handled by receive window auto-tuning, which continuously determines the optimal window size on a per-connection basis using real-time bandwidth calculations.
Improved receive window throughput increases network bandwidth utilization during data transfers. If all receivers are optimized for TCP data, Quality of Service (QoS) can help reduce congestion for networks operating at or near capacity.
Quality of Service (abbreviated QoS) refers to the ability to shape and control the characteristics of ongoing network communications services. This idea operates on the notion that transmission and error rates (along with other traffic characteristics) can be measured, improved, and guaranteed — to some extent, anyway.
Compound TCP
The Next Generation TCP/IP network stack also treats connections with large receive window sizes and large bandwidth delays to Compound TCP (CTCP), a function that aggressively increases the amount of data sent in real-time by monitoring current traffic conditions.
CTCP also ensures that it doesn’t negatively impact other existing TCP connections and complements receive window auto-tuning support to provide substantial performance gains appreciable in any high-delay, high-throughput network environment.
Explicit Congestion Notification support
Lost TCP segments are assumed to be lost, probably owing to router congestion, which triggers a congestion control mechanism that dramatically reduces a TCP sender’s transmission rate. With Explicit Congestion Notification (ECN; see RFC 3168, which you can find at www.faqs.org/rfcs/rfc3168.html) support, both TCP peers and routers experiencing congestion accordingly mark packets they forward. On receipt of such packets, a TCP peer will scale back its transmission rate to ease congestion and reduce segment loss. Windows Server 2008 now includes core support for this protocol feature.
Quality of Service (QoS) support
Windows Server 2003 and Windows XP provide QoS functionality to applications through QoS APIs, which are leveraged to prioritize time-sensitive network data delivery functions. Windows Server 2008 and Windows Vista include new facilities for network traffic management on Windows networks so that high-priority traffic is handled first, which helps with streaming media, voice over IP, video conferencing, and other applications where quick response times are needed.
Policy-based QoS for enterprise networks allows IT staff to either prioritize or manage the send rate for outbound connections, which can be confined to applications, source/destination IPv4 or IPv6 addresses, and source/destination or a range of ports.
Enhancements for high-loss environments
The Next Generation TCP/IP stack also improves network conditions in highloss environments through several optimization features that include:
• (RFC 2582) The NewReno Modification to TCP’s Fast Recovery Algorithm: The NewReno algorithm provides faster throughput by changing the way a sender can increase its sending rate when multiple segments in a given window are lost, and the sender receives partial acknowledgement only for segments actually received.
• (RFC 2883) An Extension to Selective Acknowledgement (SACK) Option for TCP: SACK allows a receiver to determine when it has retransmitted a segment unnecessarily and adjust its behavior on-the-fly to prevent further unnecessary retransmissions. Fewer retransmissions result in more optimal overall delivery.
• (RFC 3517) A Conservative Selective Acknowledgement (SACK)-based Loss Recovery Algorithm for TCP: Windows Server 2003 and Windows XP use SACK information only to determine those TCP segments that have yet to arrive. Windows Server 2008 includes a method defined in
• RFC 3517 to use SACK information for loss recovery in the event duplicate acknowledgements are received, which is maintained on a perconnection basis by the Next Generation TCP/IP stack.
• (RFC 4138) Forward RTO-Recovery (F-RTO): Spurious retransmissions can occur as a result of increases in round trip time (RTT). The F-RTO algorithm prevents unnecessary retransmissions, particularly in wireless environments where client adapters may roam from point to point, to return quickly to normal send rates.
>>> Read more about Windows Server 2008 Enhances Networking - Offloading protocol processing <<<
Source of Information : For Dummies Windows Server 2008 For Dummies
No comments:
Post a Comment