TCP for Wireless Networks
艳照门女主角Mansi Thoppian, Anjulatha Veduru
{mansi, vanjul78}@utdallas.edu
Abstract
This document is a literature survey on “TCP for Wireless Networks”. The goal is to understand the performance issues of TCP protocol when employed in wireless networks and to analyze various existing solutions. TCP is widely accepted reliable transport layer protocol for traditional wired networks, where bit-error rate is low and the packet loss are mainly due to congestion. TCP invokes congestion control and avoidance techniques to deal with such packet loss and provides end-to-end reliable communication. But in the wireless network scenario things are different.
In wireless networks the packet loss may be due to channel noi, bandwidth asymmetry or hand off mechanisms. So using the existing TCP would lead to performance degradation. Here in this literature survey we would be dealing with issues involved in extending TCP for wireless networks and discuss the various existing solutions.
Keywords: TCP, Congestion control, Bit error rate, I-TCP, MTCP, M-TCP, METP, Split Connection, Fast retransmit, Snoop protocol, TCP-F, WTCP, Explicit notification, link level retransmissions, End-to-end protocols, TCP-New Reno, TCP-SACK, TCP- SMART, Mobile TCP.
1. Introduction
Wireless technologies are playing an increasing prominent role in the global internetwork. Due to the rapid advances in the area of wireless communications and the popularity of the Internet, the provision of packet data rvices for applications like e-mail, file transfer, web surfing, interactive remote terminal applications, mobile computing etc over wireless is gaining importance. Most of the applications make u of reliable end-to-end transport protocol such as TCP. Traditionally, TCP was designed for wireline networks, where the links are relatively reliable and hosts are stationary. In such networks the channel error rates are very low and congestion is the main cau for loss of packet or unusual delays. TCP reacts to packet loss by retransmitting the missing packet and invoking congestion control. This works well in wired networks where the primary cau of packet loss is congestion.
It is obrved that TCP when ud for wireless networks results in degraded end-to-end throughput
and sub-optimal performance [18]. This is due the communication characteristics over the wireless networks. Communication characteristics of wireless networks are low bandwidth, high bit error rates and handoffs. TCP misinterprets the packet loss due to the above reasons as due to congestion and hence invokes congestion control resulting in performance degradation.
In this literature survey we explored the various existing solutions and techniques for the improvement of performance of TCP over wireless networks. We have done a thorough study on the strengths and drawbacks of each of the solutions and have prented here.
The rest of the paper is organized in the following manner. Section 2 would explain in detail the characteristics of wireless networks & Section 3 would give TCP overview. Sections 4 & 5 describe the existing solutions and their comparisons. We prent the conclusion in Section 6.
2. Issues with Mobile Wireless Networks
Mobile wireless networks can be either cellular or ad-hoc. In general the mobile wireless networks have the following salient characteristics:
1. High Bit Error rate:
The errors rates on a wireless link are much higher than tho experienced over the links in the wired network. Higher bit error rates over a wireless link are due to a combination of factors such as multi-path fading, terrain and environmental factors and interference from other transmissions.
2. Low Bandwidth:
Bandwidth is a scarce resource in ca of wireless networks. Wireless links would continue to have significantly lower bandwidth capacity than its wireline counterparts. For
< compare the bandwidth of a typical Ethernet which is around 10 Mbps to that of Lucent
Wave LAN which is just 2Mbps.
3. Changing topology:
Wireless hosts may move frequently while communicating. In cellular model the movements can result in handoffs. Communication paus during handoffs are perceived as periods of heavy data loss by transport and higher-level protocols. In ad-hoc model such topology changes are more frequent and unpredictable. Such topology changes lead to route disruptions.
The communication characteristics put together contribute to the vere degradation in TCP performance over such networks.
3. TCP Overview [1]
TCP achieves reliability by requiring the nder to retransmit lost packets. For this, the receiver would nd acknowledgements to the nder on receipt of packets. The acknowledgements can be cumulative.
TCP nder maintains a congestion window that determines the maximum amount of unacknowledged data that is already nt. Whenever the nder detects loss it would drastically reduce the congestion window size and thereby reduce the amount of data nt by the nder in the round trip time (RTT). It exponentially backs off the retransmission timer and enters the slow start recovery pha.
Sender can detect loss of packet using one of following mechanisms:
Sender starts a retransmission timer when it nds the packet. If the acknowledgement is not received before the timeout then the packet is presumed to be lost.
最简单快速去除头皮屑
Some implementations of TCP u fast retransmit approach in which the receiver on receiving each out-of-order packet nds duplicate acknowledgement containing the quence number of packet it is expecting. The nder on receiving three such duplicate acknowledgements with same quence number presumes the packet with that quence number is lost.
4. Existing Solutions
Several schemes have been propod to improve TCP performance over wireless networks. Most of the solutions were developed keeping the following model in mind:
茶的英文
There is a fixed host (mostly a ba station), which is connected to the wired network on one end and to the wireless network on the other end.
The solutions can mainly be classified into the following categories:
Split connection protocols:
The split connection approach requires the connection between the fixed host and the mobile host to be split into two parate connections at the ba station – one connection between the nder and the ba station and the other between the ba station and the receiver. The solutions such as I-TC
P [5], MTCP [4], M-TCP [9] & METP [14] come under this category.
End-To-End protocols:
In an end-to-end approach the nder and receiver adapt a mechanism to handle all possible packet loss. The solutions such as TCP New Reno [2], Fast Retransmission [6], Selective Acknowledgement [3][11], SMART-bad Acknowledgement [11][12] and WTCP for WWANs [22] u this approach.
Link layer protocols:
乘法交换律Link layer protocols implement their own link level retransmission techniques at the wireless links. In addition to this, TCP implements its own end-to-end retransmission protocol. The solutions such as standard Link Level Retransmissions [11], Link Level Retransmissions with SMART – bad [11][12], Snoop protocol [7][8], Link Level TCP- Aware [11], Delayed Duplicate Acknowledgements [17], Link Level SMART with TCP - Aware [11][12] and WTCP [13] fall under this category.
Explicit Notification protocols:
革命歌曲Explicit notification protocols explicitly notify the cau for the packet loss to the nder.
永远造句The solutions such as Explicit Loss Notification [11][15], Explicit Congestion Notification
[19], Feedback Bad Scheme (TCP-F) [10] and Mobile TCP [16] fall under this category. 4.1. Split Connection Protocols
In split connection approach the connection between the mobile host and the fixed host is split into two connections- one between the mobile host and its ba station over the wireless medium and another between the ba station and the fixed host over the fixed network. Thus the protocols shield the wired network from the uncertainties of the wireless networks by having two parate connections. They impo incread complexity and dependency on the ba station leaving the mobile host simple. For example, the fixed host will e failure of a ba station as failure in the mobile host. It also requires the ba station to have large buffers to buffer the gments destined for the mobile hosts.
The following solutions fall under this category:
4.1.1. I-TCP [5]
Indirect-TCP is bad on split connection approach. The TCP connection to the fixed host is actually
made by the ba station on behalf of the mobile host. When the mobile host requests an I-TCP connection with a fixed host the ba station of the cell containing the mobile host establishes a socket with the mobile host’s address and port number. It also opens another socket with its own address and some port to communicate with the mobile host over the wireless link. The packets nt to the mobile host are first received and buffered at the ba station. Ba station
家长评价怎么写>天人合一
acknowledges the receipt of packets to the fixed host before forwarding the packets to the mobile host over the wireless link. During the handoffs, the new ba station creates two connections corresponding to the I-TCP connection with the same endpoint parameters that the sockets at the old ba station have associated with them.
• Analysis:
The end-to-end TCP mantics is not prerved in this approach. It achieves better throughputs than standard TCP only when the disconnections are not lengthy. If there are frequent handoffs then the overhead involved in the connection state transfer between the old and the new ba stations may be large and hence increa the handoff latency. The ba station needs to maintain large buffers to ensure that the data can be buffered during lengthy disconnections without leading to buffer overflow.
4.1.2. Multiple TCP (MTCP) [4]
MTCP is also bad on split connection approach. In this protocol a ssion layer protocol (Mobile host protocol) is added at the ba station and the mobile host. This ssion layer protocol establishes two TCP connections - one over the wired path and the other over the wireless link. The ssion layer at the mobile host intercepts a TCP connection request from the mobile host to the fixed host and establishes a TCP connection with its peer at the ba station. The ssion layer at the ba station ts up a ssion layer agent on behalf of the requested connection. This agent in turn establishes the connection to the fixed host on behalf of the mobile host. This agent relays the traffic from the first connection to the cond connection.
Similar to I-TCP here also the ba station buffers the data nt to the mobile host from the fixed host. It gments the buffered data into smaller gments and forwards it to the mobile host.
During handoffs the old ba station nds the complete state information to the new ba station and the new ba station’s ssion layer takes over the connection.
Two implementations for ssion layer protocols are propod: In the first, MHP (Mobile host protocol), the connection over the wireless link is TCP. In the cond, the SRP (Selective Repeat Pro
tocol) is ud over the wireless link.
This approach is quite similar to I-TCP except that it introduces a ssion layer protocol to manage the two connections.
4.1.3. M-TCP [9]
The split connection approach is ud in implementing M-TCP while maintaining end-to-end TCP mantics. This approach us the mobile network architecture as the network model. In mobile network architecture, the Mobile Hosts (MH) communicate with Mobile Support Station (MSS/ba stations) nodes in each cell. Several MSSs are controlled by a single Supervisor Host (SH). The SH is connected to the fixed network. SH maintains connections for the mobile urs, handles flow-control and is responsible for maintaining the negotiated quality of rvice.
In M-TCP the transport connection is split into two connections at the SH. The TCP client between the nder at the fixed host and the SH is the SH-TCP which handles the connection between the fixed host and SH. The other TCP client between the SH and the MH is the M-TCP client which handles the connection between SH and MH. When SH-TCP client receives a gment from the TCP nder, it pass the gment to the M-TCP client. SH-TCP client is notified of MH ACKs by the
M-TCP client running at the SH. SH-TCP client on receiving ACKs forwards the ACKs to the nder.
Let W denote the currently advertid receiver window size at the SH-TCP. Say the window contains w<=W bytes. When the MH has ACK’ed bytes up to w1<=w, then SH-TCP nds ACKs for bytes up to w1-1. When MH ACKs more data, more ACKs are generated but one last
byte is always left unacknowledged. When a MH disconnects after acknowledging w1 bytes then M-TCP assumes that the MH has been temporarily disconnected becau it stops receiving ACKs. The M-TCP nds an indication of this fact to the SH-TCP who then nds the deferred ACK for the w1 byte to the nder and the window size as zero. The nder on receiving this ACK goes into persist state. In this state the TCP nder will not suffer from retransmit timeouts and will not exponentially back off its retransmission timer, nor will it clo its congestion window.
During the whole disconnection duration, the SH-TCP keeps nding ACKs for the persist packets nt by the TCP source to keep the connection alive. When the MH regains its connection, it nds a greeting packet to the SH. M-TCP is notified of this event and it pass on this information to SH-TCP which, in turn, nds an ACK to the nder and reopens its receive window (and hence the nder’s transmit window). This allows the nder to leave the persist mode and begin nding data
again. Now the nder can transmit at full speed, as it never performed congestion control or slow-start mechanisms.
The same strategy is applied when the MH has very little available bandwidth. SH-TCP still nds an ACK for byte w1 with the nder window size t to 0. SH-TCP estimates the round trip time to the TCP nder and estimates RTO interval. It us this information to preemptively shrink the nder’s window before the nder goes into exponential back off.
At the MH, the M-TCP is notified (by the communications hardware) that the connection to its MSS has been lost. Now the M-TCP freezes all its M-TCP timers. This ensures that disconnections do not cau the MH’s M-TCP to invoke congestion control. When the connection is regained, M-TCP at the MH nds a specially marked ACK to M-TCP at the SH which contains the quence number of the highest byte received thus far. It also unfreezes the M-TCP timers to allow normal operation to resume.
• Analysis:
The end-to-end mantics of the TCP are prerved. M- TCP works well in the prence of frequent disconnection events and over low bit-rate wireless links subject to dynamically changing bandwidth.
There is no requirement for a sophisticated MSS.
Since a single SH controls veral MSSs, the roaming MH remains within the domain of the same SH for long periods of time. This makes easier to manage the MH’s connections and ensures that any MH state maintained at the SH need only be moved infrequently.
The SH becomes extremely complex as it has to now handle all connections that may originate from mobiles located in veral cells.
In situation where the nder has only limited amount of data to nd then the above approach without any modification would result in repeated nder timeouts leading to subquent disconnection. This problem is handled by not allowing SH-TCP to shrink the nder’s window size when there is “no more new ACKS” from the MH that will allow it “to open up the nder’s window backup”. Thus, when the SH-TCP thinks that the nder will timeout and if the MH is in a cell with plenty of available bandwidth, it simply ACKs the last byte. At this point, there will not be any saved bytes at SH-TCP to allow it to shrink the nder’s window. There is no problem here becau the nder is not having any new gments to nd to the MH. As and when the nder does transmit new data to the MH, SH-TCP reverts to its previous behavior described above. However, here it is assumed that SH-TCP has enough knowledge about the nder’s retransmission timeout period.
It is assumed that link layer solutions will assure the high bit error rates en in mobile environment.
This protocol performs well in situations where the nder has a continuous stream of data to nd. In the circumstances disconnections due to hand-offs are handled efficiently.
4.1.4. Mobile-end TCP (METP) [14]
The Mobile-end protocol hides the wireless link loss from the nder by replacing the TCP/IP over the wireless link by a simple protocol with smaller headers if the link is the last hop along a data path. This protocol exploits the link-layer acknowledgements and retransmissions to quickly recover loss over the wireless link.
The communication between the ba station and the mobile host rembles that between an application process and a transport layer protocol within a usual machine. This protocol tries to take advantage of the fact that the hop between a mobile host and its ba station is the first or last one along a data path. Here the mobile hosts do not perform datagram forwarding. Only part of the IP functionalities is shifted to the ba station, which are handled by METP at the ba station.
METP at the ba station accepts IP datagrams destined for the mobile host as if they were destined
for it. It strips the datagram of its IP header and delivers it to the higher layer, since the transport layer of the mobile host is also shifted to the ba station. Re-asmbly of IP fragments is also done at the ba station. The header checksum (and similarly any higher layer checksum) is replaced by the link layer CRC since there is only one hop between the mobile host and the ba station.
All TCP connections are handled at the ba station by METP on behalf of the mobile host. METP acts as a proxy transport protocol and keeps intact all the interfaces traditionally handled by the TCP/IP stack. METP negotiates with another host in the fixed network to open or clo a TCP connection, possibly with a request from the mobile host, and keeps the connection state and nding and receiving buffers.
When the mobile host has data to nd through the TCP connection, it actually nds the data to the ba station, which puts them in nding buffer of the connection for the METP to nd out the TCP gment to the destination. A parate process tries to nd data in the receiving buffer to the mobile host. Likewi, when a TCP gment destined for a mobile host arrives at the ba station, METP puts it in the receiving buffer and nds as acknowledgement back to the source. • Analysis:
The TCP end-to-end mantics are not prerved. It depends on Link layer acknowledgements and retransmissions to recover from loss due to transmission errors.
With the above approach it is possible for packets to get dropped at the receiver if the buffer is full. Such loss are avoided by the periodic transmission of feedback packet by the METP to inform the nder of how much buffer space is currently available. Then the nder will transmit data only if the receiver buffer will not overflow.
Since the wireless link is now hidden from the fixed network, small header for packets exchanged between the mobile host and the ba station is ud. The only information retained from the header is the IP address and port numbers of the source and destination. Also using header compression techniques can further reduce header size.
METP performs well in situations where the ba stations are well connected and there are fewer connections on a mobile host at any time.
4.2. Link Layer Solutions
This category of protocols hide link related loss from the TCP nder by using reliable local retransmissions. The local retransmissions u techniques that are tuned to the characteristics of the wireless link to provide significant increa in performance. The wireless link between the ba station and the mobile host implements this protocol.
The link-layer protocols operate independently of the higher layer protocols and fit well into the layered structure of network protocols.
The link-layer protocols affect TCP’s performance for two main reasons:
