Source string Source string

1295/10000 · 1295
English Actions
By combining different answers to these questions, network researchers have developed different packet discard mechanisms.
`Tail drop` is the simplest packet discard technique. When a buffer is full, the arriving packet is discarded. `Tail drop` can be easily implemented. This is, by far, the most widely used packet discard mechanism. However, it suffers from two important drawbacks. First, since `tail drop` discards packets only when the buffer is full, buffers tend to be congested and real-time applications may suffer from increased delays. Second, `tail drop` is blind when it discards a packet. It may discard a packet from a low bandwidth interactive flow while most of the buffer is used by large file transfers.
`Drop from front` is an alternative packet discard technique. Instead of removing the arriving packet, it removes the packet that was at the head of the queue. Discarding this packet instead of the arriving one can have two advantages. First, it already stayed a long time in the buffer. Second, hosts should be able to detect the loss (and thus the congestion) earlier.
`Probabilistic drop`. Various random drop techniques have been proposed. A frequently cited technique is `Random Early Discard` (RED) [FJ1993]_. RED measures the average buffer occupancy and discards packets with a given probability when this average occupancy is too high. Compared to `tail drop` and `drop from front`, an advantage of `RED` is that thanks to the probabilistic drops, packets should be discarded from different flows in proportion of their bandwidth.
Discarding packets is a frequent reaction to network congestion. Unfortunately, discarding packets is not optimal since a packet which is discarded on a network node has already consumed resources on the upstream nodes. There are other ways for the network to inform the end hosts of the current congestion level. A first solution is to mark the packets when a node is congested. Several networking technologies have relied on this kind of packet marking.
In datagram networks, `Forward Explicit Congestion Notification` (FECN) can be used. One field of the packet header, typically one bit, is used to indicate congestion. When a host sends a packet, the congestion bit is unset. If the packet passes through a congested node, the congestion bit is set. The destination can then determine the current congestion level by measuring the fraction of the packets that it received with the congestion bit set. It may then return this information to the sending host to allow it to adapt its retransmission rate. Compared to packet discarding, the main advantage of FECN is that hosts can detect congestion explicitly without having to rely on packet losses.
In virtual circuit networks, packet marking can be improved if the return packets follow the reverse path of the forward packets. It this case, a network node can detect congestion on the forward path (e.g. due to the size of its buffer), but mark the packets on the return path. Marking the return packets (e.g. the acknowledgments used by reliable protocols) provides a faster feedback to the sending hosts compared to FECN. This technique is usually called `Backward Explicit Congestion Notification (BECN)`.
If the packet header does not contain any bit in the header to represent the current congestion level, an alternative is to allow the network nodes to send a control packet to the source to indicate the current congestion level. Some networking technologies use such control packets to explicitly regulate the transmission rate of sources. However, their usage is mainly restricted to small networks. In large networks, network nodes usually avoid using such control packets. These control packets are even considered to be dangerous in some networks. First, using them increases the network load when the network is congested. Second, while network nodes are optimized to forward packets, they are usually pretty slow at creating new packets.
Dropping and marking packets is not the only possible reaction of a router that becomes congested. A router could also selectively delay packets belonging to some flows. There are different algorithms that can be used by a router to delay packets. If the objective of the router is to fairly distribute to bandwidth of an output link among competing flows, one possibility is to organize the buffers of the router as a set of queues. For simplicity, let us assume that the router is capable of supporting a fixed number of concurrent flows, say `N`. One of the queues of the router is associated to each flow and when a packet arrives, it is placed at the tail of the corresponding queue. All the queues are controlled by a `scheduler`. A `scheduler` is an algorithm that is run each time there is an opportunity to transmit a packet on the outgoing link. Various schedulers have been proposed in the scientific literature and some are used in real routers.
A very simple scheduler is the `round-robin scheduler`. This scheduler serves all the queues in a round-robin fashion. If all flows send packets of the same size, then the round-robin scheduler fairly allocates the bandwidth among the different flows. Otherwise, it favors flows that are using larger packets. Extensions to the `round-robin scheduler` have been proposed to provide a fair distribution of the bandwidth with variable-length packets [SV1995]_ but these are outside the scope of this chapter.
Distributing the load across the network
Delays, packet discards, packet markings and control packets are the main types of information that the network can exchange with the end hosts. Discarding packets is the main action that a network node can perform if the congestion is too severe. Besides tackling congestion at each node, it is also possible to divert some traffic flows from heavily loaded links to reduce congestion. Early routing algorithms [MRR1980]_ have used delay measurements to detect congestion between network nodes and update the link weights dynamically. By reflecting the delay perceived by applications in the link weights used for the shortest paths computation, these routing algorithms managed to dynamically change the forwarding paths in reaction to congestion. However, deployment experience showed that these dynamic routing algorithms could cause oscillations and did not necessarily lower congestion. Deployed datagram networks rarely use dynamic routing algorithms, except in some wireless networks. In datagram networks, the state of the art reaction to long term congestion, i.e. congestion lasting hours, days or more, is to measure the traffic demand and then select the link weights [FRT2002]_ that allow minimizing the maximum link loads. If the congestion lasts longer, changing the weights is not sufficient anymore and the network needs to be upgraded with additional or faster links. However, in Wide Area Networks, adding new links can take months.
In virtual circuit networks, another way to manage or prevent congestion is to limit the number of circuits that use the network at any time. This technique is usually called `connection admission control`. When a host requests the creation of a new circuit in the network, it specifies the destination and in some networking technologies the required bandwidth. With this information, the network can check whether there are enough resources available to reach this particular destination. If yes, the circuit is established. If not, the request is denied and the host will have to defer the creation of its virtual circuit. `Connection admission control` schemes are widely used in the telephone networks. In these networks, a busy tone corresponds to an unavailable destination or a congested network.
In datagram networks, this technique cannot be easily used since the basic assumption of such a network is that a host can send any packet towards any destination at any time. A host does not need to request the authorization of the network to send packets towards a particular destination.
Based on the feedback received from the network, the hosts can adjust their transmission rate. We discuss in section `Congestion control` some techniques that allow hosts to react to congestion.
Another way to share the network resources is to distribute the load across multiple links. Many techniques have been designed to spread the load over the network. As an illustration, let us briefly consider how the load can be shared when accessing some content. Consider a large and popular file such as the image of a Linux distribution or the upgrade of a commercial operating system that will be downloaded by many users. There are many ways to distribute this large file. A naive solution is to place one copy of the file on a server and allow all users to download this file from the server. If the file is popular and millions of users want to download it, the server will quickly become overloaded. There are two classes of solutions that can be used to serve a large number of users. A first approach is to store the file on servers whose name is known by the clients. Before retrieving the file, each client will query the name service to obtain the address of the server. If the file is available from many servers, the name service can provide different addresses to different clients. This will automatically spread the load since different clients will download the file from different servers. Most large content providers use such a solution to distribute large files or videos.
There is another solution that allows spreading the load among many sources without relying on the name service. The popular `bittorent <https://www.bittorrent.com>`_ service is an example of this approach. With this solution, each file is divided in blocks of fixed size. To retrieve a file, a client needs to retrieve all the blocks that compose the file. However, nothing forces the client to retrieve all the blocks in sequence and from the same server. Each file is associated with metadata that indicates for each block a list of addresses of hosts that store this block. To retrieve a complete file, a client first downloads the metadata. Then, it tries to retrieve each block from one of the hosts that store the block. In practice, implementations often try to download several blocks in parallel. Once one block has been successfully downloaded, the next block can be requested. If a host is slow to provide one block or becomes unavailable, the client can contact another host listed in the metadata. Most deployments of bittorrent allow the clients to participate to the distribution of blocks. Once a client has downloaded one block, it contacts the server which stores the metadata to indicate that it can also provide this block. With this scheme, when a file is popular, its blocks are downloaded by many hosts that automatically participate in the distribution of the blocks. Thus, the number of `servers` that are capable of providing blocks from a popular file automatically increases with the file's popularity.
Now that we have provided a broad overview of the techniques that can be used to spread the load and allocate resources in the network, let us analyze two techniques in more details : Medium Access Control and Congestion control.
Medium Access Control algorithms
The common problem among Local Area Networks is how to efficiently share the available bandwidth. If two devices send a frame at the same time, the two electrical, optical or radio signals that correspond to these frames will appear at the same time on the transmission medium and a receiver will not be able to decode either frame. Such simultaneous transmissions are called `collisions`. A `collision` may involve frames transmitted by two or more devices attached to the Local Area Network. Collisions are the main cause of errors in wired Local Area Networks.
All Local Area Network technologies rely on a `Medium Access Control` algorithm to regulate the transmissions to either minimize or avoid collisions. There are two broad families of `Medium Access Control` algorithms :
`Deterministic` or `pessimistic` MAC algorithms. These algorithms assume that collisions are a very severe problem and that they must be completely avoided. These algorithms ensure that at any time, at most one device is allowed to send a frame on the LAN. This is usually achieved by using a distributed protocol which elects one device that is allowed to transmit at each time. A deterministic MAC algorithm ensures that no collision will happen, but there is some overhead in regulating the transmission of all the devices attached to the LAN.
`Stochastic` or `optimistic` MAC algorithms. These algorithms assume that collisions are part of the normal operation of a Local Area Network. They aim to minimize the number of collisions, but they do not try to avoid all collisions. Stochastic algorithms are usually easier to implement than deterministic ones.
We first discuss a simple deterministic MAC algorithm and then we describe several important optimistic algorithms, before coming back to a distributed and deterministic MAC algorithm.
Static allocation methods
A first solution to share the available resources among all the devices attached to one Local Area Network is to define, `a priori`, the distribution of the transmission resources among the different devices. If `N` devices need to share the transmission capacities of a LAN operating at `b` Mbps, each device could be allocated a bandwidth of :math:`\frac{b}{N}` Mbps.
Limited resources need to be shared in other environments than Local Area Networks. Since the first radio transmissions by `Marconi <http://en.wikipedia.org/wiki/Guglielmo_Marconi>`_ more than one century ago, many applications that exchange information through radio signals have been developed. Each radio signal is an electromagnetic wave whose power is centered around a given frequency. The radio spectrum corresponds to frequencies ranging between roughly 3 KHz and 300 GHz. Frequency allocation plans negotiated among governments reserve most frequency ranges for specific applications such as broadcast radio, broadcast television, mobile communications, aeronautical radio navigation, amateur radio, satellite, etc. Each frequency range is then subdivided into channels and each channel can be reserved for a given application, e.g. a radio broadcaster in a given region.
`Frequency Division Multiplexing` (FDM) is a static allocation scheme in which a frequency is allocated to each device attached to the shared medium. As each device uses a different transmission frequency, collisions cannot occur. In optical networks, a variant of FDM called `Wavelength Division Multiplexing` (WDM) can be used. An optical fiber can transport light at different wavelengths without interference. With WDM, a different wavelength is allocated to each of the devices that share the same optical fiber.
`Time Division Multiplexing` (TDM) is a static bandwidth allocation method that was initially defined for the telephone network. In the fixed telephone network, a voice conversation is usually transmitted as a 64 Kbps signal. Thus, a telephone conservation generates 8 KBytes per second or one byte every 125 microseconds. Telephone conversations often need to be multiplexed together on a single line. For example, in Europe, thirty 64 Kbps voice signals are multiplexed over a single 2 Mbps (E1) line. This is done by using `Time Division Multiplexing` (TDM). TDM divides the transmission opportunities into slots. In the telephone network, a slot corresponds to 125 microseconds. A position inside each slot is reserved for each voice signal. The figure below illustrates TDM on a link that is used to carry four voice conversations. The vertical lines represent the slot boundaries and the letters the different voice conversations. One byte from each voice conversation is sent during each 125 microseconds slot. The byte corresponding to a given conversation is always sent at the same position in each slot.
TDM as shown above can be completely static, i.e. the same conversations always share the link, or dynamic. In the latter case, the two endpoints of the link must exchange messages specifying which conversation uses which byte inside each slot. Thanks to these control messages, it is possible to dynamically add and remove voice conversations from a given link.
TDM and FDM are widely used in telephone networks to support fixed bandwidth conversations. Using them in Local Area Networks that support computers would probably be inefficient. Computers usually do not send information at a fixed rate. Instead, they often have an on-off behavior. During the on period, the computer tries to send at the highest possible rate, e.g. to transfer a file. During the off period, which is often much longer than the on period, the computer does not transmit any packet. Using a static allocation scheme for computers attached to a LAN would lead to huge inefficiencies, as they would only be able to transmit at :math:`\frac{1}{N}` of the total bandwidth during their on period, despite the fact that the other computers are in their off period and thus do not need to transmit any information. The dynamic MAC algorithms discussed in the remainder of this chapter aim to solve this problem.

Loading…

User avatar None

String updated in the repository

cnp3-ebook / principles/sharing — English

Another way to share the network resources is to distribute the load across multiple links. Many techniques have been designed to spread the load over the network. As an illustration, let us briefly consider how the load can be shared when accessing some content. Consider a large and popular file such as the image of a Linux distribution or the upgrade of a commercial operating system that will be downloaded by many users. There are many ways to distribute this large file. A naive solution is to place one copy of the file on a server and allow all users to download this file from the server. If the file is popular and millions of users want to download it, the server will quickly become overloaded. There are two classes of solutions that can be used to serve a large number of users. A first approach is to store the file on servers whose name is known by the clients. Before retrieving the file, each client will query the name service to obtain the address of the server. If the file is available from many servers, the name service can provide different addresses to different clients. This will automatically spread the load since different clients will download the file from different servers. Most large content providers use such a solution to distribute large files or videos.
Another way to share the network resources is to distribute the load across multiple links. Many techniques have been designed to spread the load over the network. As an illustration, let us briefly consider how the load can be shared when accessing some content. Consider a large and popular file such as the image of a Linux distribution or the upgrade of a commercial operating system that will be downloaded by many users. There are many ways to distribute this large file. A naive solution is to place one copy of the file on a server and allow all users to download this file from the server. If the file is popular and millions of users want to download it, the server will quickly become overloaded. There are two classes of solutions that can be used to serve a large number of users. A first approach is to store the file on servers whose name is known by the clients. Before retrieving the file, each client will query the name service to obtain the address of the server. If the file is available from many servers, the name service can provide different addresses to different clients. This will automatically spread the load since different clients will download the file from different servers. Most large content providers use such a solution to distribute large files or videos.
2 years ago
Browse all component changes

Glossary

English English
No related strings found in the glossary.

String information

Flags
read-only
Source string location
../../principles/sharing.rst:352
String age
2 years ago
Source string age
2 years ago
Translation file
locale/pot/principles/sharing.pot, string 53