Tcpip performance (sorting)

1) Adaptive TCP acceleration technology
The size of the transmission window, that is, the amount of data sent before receiving the response, will directly affect the TCP performance. On the contrary, the performance is proportional to the return time because the Protocol needs to ensure that the data is delivered in place (an ACK packet indicates that the data has been successfully received.

In the worst case, an endpoint will wait for the other end point to respond to the data transmission situation, thus increasing the idle time of the network. This happens when the transmission window becomes very small, but it does not accurately reflect the line speed and latency.

To make the situation more complex, TCP will adjust its window size based on the response speed. The longer the connection distance, the smaller the window. If the response speed is very slow, the TCP protocol may never select the maximum window size, which means that the full capacity of many WAN connections will never be fully utilized. Therefore, the TCP protocol may degrade the WAN performance, and even when the bandwidth is still very adequate, the performance deterioration is inevitable. Similarly, retransmission will seriously affect TCP performance. 1% packet loss may cause a maximum performance loss of 80%.

The adaptive TCP acceleration solution can solve these problems using self-regulating technologies. This technology can automatically change the behavior of terminal hosts and is completely transparent to applications, networks, and end-user computers involved in transmission. The window size is determined based on the measured available connection bandwidth, thus avoiding idle bandwidth and waste.

Adaptive TCP acceleration technology will be used in the next generation of WAN optimization devices. You only need to install one device at each endpoint of the WAN connection.

To prevent impact on other devices that share the same connection and reduce their performance, this technology can automatically adjust transmission parameters based on user-defined policies and packet loss. These policies are defined using the management software of Wan-optimized devices, and the devices can interpret and implement the policies. At the same time, this technology can select the size of the TCP Window on its own, so as to achieve the highest transmission rate and minimize the number of retransmission packets when packets are lost in the WAN connection.

Adaptive TCP acceleration technology also uses dynamic setting of large windows (within the appropriate range), selective ACK, TCP Fast Start and other technologies for the first time to improve performance. Selective ACK can respond to data reception and allow immediate retransmission of lost data, thus avoiding the latency problem caused by the TCP retransmission algorithm. TCP Fast Start is also an algorithm that accelerates the growth of the TCP sending window and enables quick use of available bandwidth.

2)TCP Fast Start
A Technique For Speeding Up Web Transfers (1998)

3) Selective ACK

4) TCP/IP frame Header Compression

5) CDMA 20001x Network Optimization
The main goal of the optimization of the CDMA 1X network is to maximize the system throughput and minimize the transmission latency, so as to give full play to the wireless transmission advantages of the CDMA 1X, to a certain extent, improve the performance of the wireless data network, ensure the implementation of its wireless data service. The optimization of the network can be divided into two aspects: the optimization of the wireless network and the optimization of the Group network.

The wireless network optimization method is similar to the IS-95CDMA network optimization, but the data transmission has higher requirements on the wireless environment, and the parameters of the forward link and the reverse link are optimized. This article focuses on the grouping network optimization of CDMA 20001x.

The main purpose of grouping network optimization is to differentiate the TCP transmission characteristics from the wired environment in the existing wireless environment, through the optimization of TCP Parameters and the adoption of TCP optimization, cache devices, etc., the wireless data transmission performance of CDMA 20001x is improved.

The CDMA 20001x group network is a computer network that adopts the TCP/IP protocol, and its transmission layer mainly adopts the TCP and UDP protocols. TCP is a connection-oriented transmission control protocol for reliable data transmission under various network conditions. UDP is a non-connection data packet protocol. From the perspective of optimization, TCP parameter settings have a great impact on data transmission performance.

TCP uses a window-based Congestion Control Mechanism. Currently, four common congestion control algorithms are available: slow start, congestion avoidance, fast retransmission, and fast recovery. These congestion control algorithms are designed to avoid unreasonable network traffic in a wired network. When running TCP/IP on a wireless link, TCP congestion control is different from Wired transmission. On the wireless link, the bit rate is significantly higher than that on the wired link due to wireless signal loss and switching, so some TCP algorithms are not ideal. When data is uploaded and transmitted over a wireless channel, latency, bandwidth oscillation, asymmetry, and other factors that have not been taken into account in the TCP algorithm during online channel transmission. The TCP parameter settings, such as TCP/IP frame header compression, selective validation, and MTU Size Optimization, will have a great impact on the performance of the CDMA 20001x network.

According to the theoretical and practical test results, the setting principles of TCP parameters are as follows: Disable the TCP/IP frame header compression function; allow selective validation function; if the TCP Data Segment error rate is too high, you can try to reduce the MTU size; allow fast retransmission after the second duplicate ACK response is received; minimize the ACK latency as much as possible, and return a response every time a piece of data is received at the receiving end. In the specific network optimization process, the TCP Parameter Optimization settings must be implemented based on the actual situation.

TCP is an end-to-end transmission control protocol because wireless and wired links have different requirements on TCP Parameters. The handling of packet loss in wired and wireless networks is also different. For example, a wired network thinks that packet loss is mainly caused by network congestion. Once packet loss occurs, TCP will reduce the sending window by half to reduce congestion. However, if packet loss occurs on the wireless network, it is more likely that the transient fading of the wireless channel is caused, and the fading may be quickly restored. In this case, if the TCP window is reduced, this will reduce the data throughput. To achieve better optimization results, you can use the following two methods to optimize the CDMA 20001x Data Group network.

Transport Layer optimization. A transmission layer-based wireless gateway device can be added between wired and wireless networks to improve data throughput and transmission efficiency by separately processing wired and wireless links.

Application Layer optimization. You can improve the efficiency of the application layer by adding a device that is optimized for the application layer by caching, data compression, protocol adjustment, and re-encoding.

In addition, because various application servers on the Internet are controlled by different content providers, different server devices and operating systems are used. To ensure the data performance of the CDMA 20001x network, the network operator must provide a complete solution. Because the packet loss rate of the wired network is much lower than that of the wireless network, you only need to apply selective confirmation at both ends of the wireless link. Specifically, a wireless gateway device is added after the PDSN of the CDMA 20001x group network. It should have the bypass function. when it encounters a fault and the load is too high, the wireless network can still provide services. The wireless gateway can divide the TCP link from the CDMA 20001x terminal to the application server into two sections. The conventional TCP mechanism is still used in the wired link, and the selective validation is adopted for the wireless side. The wireless gateway can also adjust the MTU size as needed to reduce the response delay. In addition, due to the great change in latency of the wireless network, it is easy to time out. According to the TCP congestion control mechanism, once the time-out occurs, the system restarts slowly, which reduces the transmission efficiency. The wireless gateway can accurately calculate the round-trip time of the wireless link TCP, reduce the slow start, and improve the transmission efficiency. Through the segment processing of the TCP transmission path through the wireless gateway, the data throughput of the offline network can be significantly improved.

For the CDMA 20001x network, it can also be optimized based on the application layer. By storing frequently accessed content in the cache and compressing uncompressed graphics and text files, you can increase the access speed. The HTTP Version used to access the Web page also affects the access speed. On the application optimization server, you can convert multiple connection requests from HTTP1.0 to one connection request from HTTP1.1 to improve the actual network throughput.

6) Performance Optimization

● TCP Window Size


There are many TCP parameters that affect network performance, but the most important thing is the size of the TCP window. It determines how much data a system can transmit at a given time. Each TCP packet has a header, and the header information contains a certain "window" field to indicate the size of the system's accept data buffer. A large window can allow the remote system to transmit more data, while a small window limits the amount of data transmitted, thus affecting network performance. To use the network to the maximum extent, select the appropriate window size to match the available bandwidth. If the TCP window is too small, end users will not be able to make full use of the available bandwidth on the network. If the TCP Window port is too large, it may cause errors and recovery problems, this will also greatly reduce network performance.

Generally, all systems have a default TCP window size. Some applications allow system-level APIs to set the TCP Window Size Based on each connection. It is not easy to accurately select the size of TCP Window ports suitable for the network, and many measurements and calculations are required. Of course, these calculations are for connections. Therefore, if there are many connections from different networks, you may feel at a loss, in this case, it is best to use the default settings (because the TCP window size suitable for a network may not apply to another network ). Therefore, when all users on the Internet are generally similar and all terminal nodes (including servers and clients) are set in the same way, high-speed setting truly reflects its value.


● Transfer data volume


When determining the size of the TCP window, the two most important parameters are round-trip latency and end-to-end bandwidth. These two parameters determine the amount of data online (in transmission) at any given time.

Latency refers to the time required for data from one system to another (including the online transmission time and the processing time of the two systems ). The longer the data is transmitted online, the more data can be accommodated online at any given time. Round-trip latency (including the data transmission time and the response time returned) is a more important parameter.

The same end-to-end bandwidth also affects the data volume on the Internet. If the bandwidth is 1 Mbps, 1 MB of data can be sent per second. End-to-end bandwidth is the throat of network connection. If there is a Mbps Nic and the remote system uses a 64 kbps ISDN, the maximum bandwidth for end-to-end connection is 64 kbps. In addition, the round-trip delay should include the sum of the transmission time between the two.

Multiply the round-trip latency by the end-to-end bandwidth, which is the maximum data volume on any TCP virtual circuit. This value actually determines the size of the TCP window. If a small value is set, the sending end will be limited by the size of the remote system window, and the potential of the network cannot be realized. When the TCP window is full, the sender must stop sending data until receiving the response. However, if the window size is correctly set, the sender can continuously send data, because when all the data is transmitted, the receiving response appears.

It is easy to measure the round-trip latency. ping the remote system. It is best to do it several times in normal working days and select the most common delay time. It is best to select a common package size during ping. The more devices on the Internet, the longer the delay. Therefore, the measurement network is very important.


● Maximum segment size (MSS) coefficient


Only the data volume on the internet is noticed, and the size of the TCP Window value cannot be determined. You must know the maximum size (MSS) of the connection ). Because of the TCP delay in receiving the response, the algorithm requires that the receiving response can be sent only when two complete TCP packets are received.

In fact, the default window size should be 4 times that of MSS. If the receiving window is twice that of the MSS, the sender must wait for the receiving response. If the receiving window is four times that of the MSS, the sender can send at least four packets, when the first two packets receive a response, the last packet is just sent. If the network latency is long, the size of the Receiving Window should be 6 times or 8 times that of the MSS.

Generally, 10 Mbps Ethernet with a delay of 2 ms should use a window size of 4 times that of MSS. Otherwise, this value will greatly affect network performance.

When the window is too large, it is difficult for TCP to restore the lost data. If a remote Web server learns that the window size of a customer is small kb, it will try to transmit small kb even if the data that can be transmitted on the Internet is only 4 kb at the same time, in this way, the data in kilobytes will be queued at a vro between the server and the client. If data is lost and needs to be re-transmitted, the data to be re-transmitted must be placed behind it. As a result, the client considers the connection to be unrecoverable and will discard the connection. In addition, the client repeatedly sends responses to the server, which may also cause the server to interrupt the connection. Sometimes to make the window more efficient, you may need a mobile server to change the network structure.

The TCP window has only 16 bits, so the maximum value is 65535. If the network transmission speed is slow but there is a large capacity, the optimized window size may exceed it. In this case, you need to use the advanced options of the TCP window.