4.8 Active Queue Management (Active Queue Management)
As mentioned earlier, the TCP responds to congestion is to close the window and call slowly. Networks of delayed time are recovered in this situation also require a particularly long time, so the generation of congestion must be avoided in the remote narrowband network (LTNS). To this end, the activity queue management technology is made as a performance enhancement of the router across the Internet [RED]. The main way of this type of configuration mechanism prevents 'congestion collapse' by controlling the average queue size at the router. When the length of this average queue is increased, random early detection (Random Early Detection [Red] adds the chance of discarding the bad bag.
The benefits are:
- Reduce the number of packets of the router. The RED mechanism will take the initiative to discard a small amount of bags before generating severe congestion, and avoid the surge in packet loss. It is said that its purpose is to avoid discard N packets (M less than n) later by discarding the M packets early.
- Reduce delay. Implemented through smaller queues, this is especially important for interactive applications. For interactive applications for wireless connections, its inherent delay has reached the tolerance limit of users.
- Avoid death. Since the lack of router resources (or loss of corresponding packets) may cause a connection to swallow. After the activity queue management technology, the chance of finding the corresponding processing buffer in the router will also increase.
The activity queue management has two parts: First, the router is detected by the congestion before its resource depleting; second, it provides some congestion indications. The package discarded by RED detection is only an example of the latter. Other ways to congestion is to use ECN [ECN], "as described above" to detect bad pack loss by explicit notification ".
Recommendation: The current RED has been used on the Internet, and the LTN should also follow execution. The ECN configuration should make up the deficiencies of Red.
4.9 Time Order Algorithm (Scheduling Algorithms)
The activity queue management helps the control queue length. Moreover, universal solutions require other scheduling algorithms to replace FIFO, thereby reaching the following performance:
Montenegro, et al. Informational [Page 21]
1. Fair (realized by control bandwidth, that is, different types of packet flows can only use the bandwidth available)
2. Throughput (achieved by increasing the utilization of the transfer of the wireless channel)
For example, in a batch transmission session (Bulk Transfer Sessions) and interactive applications (such as Telnet or web browsing), fairness is particularly important.
Suggestions in this regard are:
- Fair queueing, FQ) [Demers90]
- Class-based Queueing, CBQ) [FLOYD95]
Even if they are only partially implemented in the field of wireless communications, these recommendations are still very charming for wireless LTN environments because they will occupy all available bandwidth after batch TCP transmission starts, thereby causing difficulties to interact with interactive applications.
In the basic architecture described earlier, mobile devices typically communicate directly with a wireless connection point (ie, intermediate node) at a given time. In some wide area network (W-WAN), it is possible to directly locate other mobile devices directly with the same unit (Cell). Direct communication with each such wireless connection point may have to cross the path, each path may exhibit a unique wireless connection feature. Channel State Dependent Packet Scheduling, CSDP [BBKT96] can track the state of a wide variety of wireless connections (determined by the target device), and provide a priority to the "good (good" stateful wireless connection Take care. This avoids sending or waiting for response to "bad (BAD)" wireless connection point, thereby increasing throughput. The better suggestion of this envision is to unite CSDP [FSS98] and wireless-enhance CBQ to achieve the purpose of improving fairness and throughput.
Suggestion: There is no suggestion, there is still a further study.
4.10 Split TCP and Performance-Enhange PROXIES (PEPS))
It can be used in a relatively vivid way to compare the different wireless connections and wired connections. A very common approach is to apply some impedance matching at the intermediate medium nodes in two different techniques.
Montenegro, et al. Informational [Page 22]
This idea is to replace the end-to-to-end TCP connection with two distinctive connections: one is through a wireless connection, and the other is through the wired portion corresponding to it. Both methods can make TCP sessions in very different network features, and can choose the most suitable policy based on special requirements of the media. For example, in the case of a special LTN topology, the Fast Retransmit will be retransmitted after receiving the first repetition; five recovery, the RTT connected to the LTN At particular long, the congestion window is not reduced, and this purpose, the ideal method is not to record the packet, thereby avoiding congestion. In addition, the long-delayed connection or connection is higher than the relative bandwidth delay, which may be better with a "slow start" manner with relatively initialization windows, which even more than four segments. These various forms of TCP improvements should be applied to the LTN connection, but they need to discuss, but they cannot be used to configure end-to-end connections across Internet. In the LTN topology, since the fundamental characteristics of the connection are already clear, there is a variety of similar performance enhancement schemes in the premise of not endangering the global Internet.
In some suggestions, in addition to the PEP mechanism at the intermediate media node, there are some conventional agreements, such as [WAP], [YB94], [MOWGLI], etc.
Even if you can get the same or better characteristics by using a non-TCP protocol, the value of the wireless TCP optimization research and the compatibility with the Internet are still the main reason for using TCP connections on wireless connections (enhanced recommendations to see the fifth section below. ).
4.10.1 Split TCP (Split TCP Approaches)
Split TCP (SPLIT-TCP) It is recommended to include I-TCP [ITCP] and MTCP [YB94], which are improved by the discarding end peer syntax. The MOWGLI system [MOWGLI] is recommended to provide enhancement support for segmentation methods in all protocol layers, not just transport layers. MowGli provides an option that can be customized, specifically for LTN connection protocols [Krlka97] to replace TCP / IP kernel protocols in the original LTN connection. In addition, the protocol also provides a number of useful features for the LTN network, for example, it provides multiple techniques based on priority concurrent connection and shared flow control, even in the case of a background transmission of the affected bandwidth, can still be interactive Timely delivery of the type of connection is guaranteed. Similarly, under this option, MOWGLI retains the socket syntax on the mobile device so that the original application can continue to be run without modifying.
Montenegro, et al. Informational [Page 23]
The method of splitting TCP also has its advantages. The advantages related to it are:
- Split the end-to-end TCP connection into two parts, so that the problem in the wireless connection does not associate with the cable Internet path, and vice versa, thus, the split TCP method supports the use of local solutions to solve local problems in the wireless connection . For example, it automatically distinguishes the packet loss caused by the wired Internet congestion related packet loss and wireless connection, as if these are the same in separated TCP connections. Even if the two network segments (wired and wireless, translator's note) simultaneously generate congestion, it is still possible to solve the same method due to different characteristics. In addition, temporary disconnections in wireless connections can be effectively isolated from the wired Internet.
- When a TCP connection is just a few-level wireless connection, all or part of the connection characteristics of the wireless TCP path can be known. For example, for a specified connection, we may know that the connection can provide a reliable package transfer capability, and the order of the packet will not be chaotic, and the connection will not be plagued by congestion. With these information related to TCP channels, the TCP Mitigations program will be a relaxed task. In addition, there are several weight loss schemes that are unable to be safe for global Internet, but can be successfully applied to wireless connections.
- After dividing a TCP connection into two separate parts, you can apply many recently, and apply the recommendations of TCP performance in wireless connections. Because only need to modify the TCP implementation of the mobile device and the intermediate media node, Many Internet hosts can continue to run traditional TCP implementations without having to make any modifications. Any weight loss scheme that needs to be modified to modify the cable host TCP, to be promoted is almost outless.
- Allowing performance enhancements using multiple application levels to increase significant performance (see Section 4.10.2).
The disadvantage of the segmentation TCP scheme is as follows:
- The main criticism of the split TCP program is because it breaks the TCP end-to-end syntax, which will bring many more serious problems, which will result in the end-to-end application of the IP layer security mechanism, Make IPSEC-based applications cannot obtain end-to-end security.
Montenegro, et al. Informational [Page 24]
Although IPsec can apply two partial independent applications, the intermediate media node is required to serve the security link between mobile devices and remote hosts, which is not good to accept in many cases. At this time, other security mechanisms on the transmission layer are required to implement end-to-end security, such as TLS [RFC2246] or SOCKS [RFC1928]. - Another disadvantage of breaking the end-to-end syntax is that the collision generated at the intermediate media node will become unable to recover, resulting in the abortment of the TCP connection. This should be considered a serious problem to see the frequency generated by this conflict.
- There are many opportunities to prove that when the TCP end-to-end syntax is broken, the application that reliably provides reliable data transfer will be more vulnerable. About "design good applications will never depend all on TCP to ensure that the end-to-end reliability in its application level is just exaggerated. First, the current TCP APIs, such as the Berkeley Socket interface, the application cannot know when the recently sent user data is confirmed to the TCP sender; second, even if the application gets TCP confirmation, the sender application is still unable to I know if the receiving program has received the data: it only knows that the data has reached the receiving buffer of the TCP receiver; The data has made a corresponding action.
- When the mobile device moves, it obeys a handover between its service base station. If the base station is a split TCP connection intermediate node, the TCP state at both ends of the previous intermediate node must also be passed to the new intermediate node to confirm the continuity of this split TCP connection operation. This requires some additional work and can trigger overload. However, for the vast majority of broadband WAN (W-WAN), unlike broadband local area network (W-LAN), the W-WAN base station does not provide connection points between mobile devices and wired Internet (such base stations may also connect IP) There is no stack). As an alternative, when the mobile device moves, W-WAN should maintain this type of motor and keep the mobile device unchanged with the connection point of the mobile device. Thus, in most W-WANs, the transfer of TCP status is not necessary.
- The packet through all the protocol layers above the transport layer and the link layer, which will cause additional loads at the intermediate node. In a low-bandwidth LTN network, this overload does not cause serious performance problems such as high bandwidth W-WAN (it is possible to trigger performance issues, translator's note).
Montenegro, et al. Informational [Page 25]
