Source string Source string

741/10000 · 741
English Actions
A network is designed to support a potentially large number of users that exchange information with each other. These users produce and consume information which is exchanged through the network. To support its users, a network uses several types of resources. It is important to keep in mind the different resources that are shared inside the network.
The first and more important resource inside a network is the link bandwidth. There are two situations where link bandwidth needs to be shared between different users. The first situation is when several hosts are attached to the same physical link. This situation mainly occurs in Local Area Networks (LAN). A LAN is a network that efficiently interconnects several hosts (usually a few dozens to a few hundreds) in the same room, building or campus. Consider for example a network with five hosts. Any of these hosts needs to be able to exchange information with any of the other hosts. A first organization for this LAN is the full-mesh.
The full-mesh is the most reliable and highest performing network to interconnect these five hosts. However, this network organization has two important drawbacks. First, if a network contains `n` hosts, then :math:`\frac{n\times(n-1)}{2}` links are required. If the network contains more than a few hosts, it becomes impossible to lay down the required physical links. Second, if the network contains `n` hosts, then each host must have :math:`n-1` interfaces to terminate :math:`n-1` links. This is beyond the capabilities of most hosts. Furthermore, if a new host is added to the network, new links have to be laid down and one interface has to be added to each participating host. However, full-mesh has the advantage of providing the lowest delay between the hosts and the best resiliency against link failures. In practice, full-mesh networks are rarely used except when there are few network nodes and resiliency is key.
The second possible physical organization, which is also used inside computers to connect different extension cards, is the bus. In a bus network, all hosts are attached to a shared medium, usually a cable through a single interface. When one host sends an electrical signal on the bus, the signal is received by all hosts attached to the bus. A drawback of bus-based networks is that if the bus is physically cut, then the network is split into two isolated networks. For this reason, bus-based networks are sometimes considered to be difficult to operate and maintain, especially when the cable is long and there are many places where it can break. Such a bus-based topology was used in early Ethernet networks.
A third organization of a computer network is a star topology. In such networks, hosts have a single physical interface and there is one physical link between each host and the center of the star. The node at the center of the star can be either a piece of equipment that amplifies an electrical signal, or an active device, such as a piece of equipment that understands the format of the messages exchanged through the network. Of course, the failure of the central node implies the failure of the network. However, if one physical link fails (e.g. because the cable has been cut), then only one node is disconnected from the network. In practice, star-shaped networks are easier to operate and maintain than bus-shaped networks. Many network administrators also appreciate the fact that they can control the network from a central point. Administered from a Web interface, or through a console-like connection, the center of the star is a useful point of control (enabling or disabling devices) and an excellent observation point (usage statistics).
A fourth physical organization of a network is the ring topology. Like the bus organization, each host has a single physical interface connecting it to the ring. Any signal sent by a host on the ring will be received by all hosts attached to the ring. From a redundancy point of view, a single ring is not the best solution, as the signal only travels in one direction on the ring; thus if one of the links composing the ring is cut, the entire network fails. In practice, such rings have been used in local area networks, but are now often replaced by star-shaped networks. In metropolitan networks, rings are often used to interconnect multiple locations. In this case, two parallel links, composed of different cables, are often used for redundancy. With such a dual ring, when one ring fails all the traffic can be quickly switched to the other ring.
A fifth physical organization of a network is the tree. Such networks are typically used when a large number of customers must be connected in a very cost-effective manner. Cable TV networks are often organized as trees.
Sharing bandwidth
In all these networks, except the full-mesh, the link bandwidth is shared among all connected hosts. Various algorithms have been proposed and are used to efficiently share the access to this resource. We explain several of them in the Medium Access Control section below.
Fairness in computer networks
Sharing resources is important to ensure that the network efficiently serves its user. In practice, there are many ways to share resources. Some resource sharing schemes consider that some users are more important than others and should obtain more resources. For example, on the roads, police cars and ambulances have priority. In some cities, traffic lanes are reserved for buses to promote public services, ... In computer networks, the same problem arise. Given that resources are limited, the network needs to enable users to efficiently share them. Before designing an efficient resource sharing scheme, one needs to first formalize its objectives. In computer networks, the most popular objective for resource sharing schemes is that they must be `fair`. In a simple situation, for example two hosts using a shared 2 Mbps link, the sharing scheme should allocate the same bandwidth to each user, in this case 1 Mbps. However, in a large networks, simply dividing the available resources by the number of users is not sufficient. Consider the network shown in the figure below where `A1` sends data to `A2`, `B1` to `B2`, ... In this network, how should we divide the bandwidth among the different flows ? A first approach would be to allocate the same bandwidth to each flow. In this case, each flow would obtain 5 Mbps and the link between `R2` and `R3` would not be fully loaded. Another approach would be to allocate 10 Mbps to `A1-A2`, 20 Mbps to `C1-C2` and nothing to `B1-B2`. This is clearly unfair.
In large networks, fairness is always a compromise. The most widely used definition of fairness is the `max-min fairness`. A bandwidth allocation in a network is said to be `max-min fair` if it is such that it is impossible to allocate more bandwidth to one of the flows without reducing the bandwidth of a flow that already has a smaller allocation than the flow that we want to increase. If the network is completely known, it is possible to derive a `max-min fair` allocation as follows. Initially, all flows have a null bandwidth and they are placed in the candidate set. The bandwidth allocation of all flows in the candidate set is increased until one link becomes congested. At this point, the flows that use the congested link have reached their maximum allocation. They are removed from the candidate set and the process continues until the candidate set becomes empty.
In the above network, the allocation of all flows would grow until `A1-A2` and `B1-B2` reach 5 Mbps. At this point, link `R1-R2` becomes congested and these two flows have reached their maximum. The allocation for flow `C1-C2` can increase until reaching 15 Mbps. At this point, link `R2-R3` is congested. To increase the bandwidth allocated to `C1-C2`, one would need to reduce the allocation to flow `B1-B2`. Similarly, the only way to increase the allocation to flow `B1-B2` would require a decrease of the allocation to `A1-A2`.
Network congestion
Sharing bandwidth among the hosts directly attached to a link is not the only sharing problem that occurs in computer networks. To understand the general problem, let us consider a very simple network which contains only point-to-point links. This network contains three hosts and two routers. All the links inside the network have the same capacity. For example, let us assume that all links have a bandwidth of 1000 bits per second and that the hosts send packets containing exactly one thousand bits.
In the network above, consider the case where host `A` is transmitting packets to destination `C`. `A` can send one packet per second and its packets will be delivered to `C`. Now, let us explore what happens when host `B` also starts to transmit a packet. Node `R1` will receive two packets that must be forwarded to `R2`. Unfortunately, due to the limited bandwidth on the `R1-R2` link, only one of these two packets can be transmitted. The outcome of the second packet will depend on the available buffers on `R1`. If `R1` has one available buffer, it could store the packet that has not been transmitted on the `R1-R2` link until the link becomes available. If `R1` does not have available buffers, then the packet needs to be discarded.
Besides the link bandwidth, the buffers on the network nodes are the second type of resource that needs to be shared inside the network. The node buffers play an important role in the operation of the network because that can be used to absorb transient traffic peaks. Consider again the example above. Assume that on average host `A` and host `B` send a group of three packets every ten seconds. Their combined transmission rate (0.6 packets per second) is, on average, lower than the network capacity (1 packet per second). However, if they both start to transmit at the same time, node `R1` will have to absorb a burst of packets. This burst of packets is a small `network congestion`. We will say that a network is congested, when the sum of the traffic demand from the hosts is larger than the network capacity :math:`\sum{demand}>capacity`. This `network congestion` problem is one of the most difficult resource sharing problem in computer networks. `Congestion` occurs in almost all networks. Minimizing the amount of congestion is a key objective for many network operators. In most cases, they will have to accept transient congestion, i.e. congestion lasting a few seconds or perhaps minutes, but will want to prevent congestion that lasts days or months. For this, they can rely on a wide range of solutions. We briefly present some of these in the paragraphs below.
If `R1` has enough buffers, it will be able to absorb the load without having to discard packets. The packets sent by hosts `A` and `B` will reach their final destination `C`, but will experience a longer delay than when they are transmitting alone. The amount of buffering on the network node is the first parameter that a network operator can tune to control congestion inside his network. Given the decreasing cost of memory, one could be tempted to put as many buffers [#fbufferbloat]_ as possible on the network nodes. Let us consider this case in the network above and assume that `R1` has infinite buffers. Assume now that hosts `A` and `B` try to transmit a file that corresponds to one thousand packets each. Both are using a reliable protocol that relies on go-back-n to recover from transmission errors. The transmission starts and packets start to accumulate in `R1`'s buffers. The presence of these packets in the buffers increases the delay between the transmission of a packet by `A` and the return of the corresponding acknowledgment. Given the increasing delay, host `A` (and `B` as well) will consider that some of the packets that it sent have been lost. These packets will be retransmitted and will enter the buffers of `R1`. The occupancy of the buffers of `R1` will continue to increase and the delays as well. This will cause new retransmissions, ... In the end, only one file will be delivered (very slowly) to the destination, but the link `R1-R2` will transfer much more bytes than the size of the file due to the multiple copies of the same packets. This is known as the `congestion collapse` problem :rfc:`896`. Congestion collapse is the nightmare for network operators. When it happens, the network carries packets without delivering useful data to the end users.
Congestion collapse on the Internet
Congestion collapse is unfortunately not only an academic experience. Van Jacobson reports in [Jacobson1988]_ one of these events that affected him while he was working at the Lawrence Berkeley Laboratory (LBL). LBL was two network nodes away from the University of California in Berkeley. At that time, the link between the two sites had a bandwidth of 32 Kbps, but some hosts were already attached to 10 Mbps LANs. "In October 1986, the data throughput from LBL to UC Berkeley ... dropped from 32 Kbps to 40 bps. We were fascinated by this sudden factor-of-thousand drop in bandwidth and embarked on an investigation of why things had gotten so bad." This work lead to the development of various congestion control techniques that have allowed the Internet to continue to grow without experiencing widespread congestion collapse events.
Besides bandwidth and memory, a third resource that needs to be shared inside a network is the (packet) processing capacity. To forward a packet, a router needs bandwidth on the outgoing link, but it also needs to analyze the packet header to perform a lookup inside its forwarding table. Performing these lookup operations require resources such as CPU cycles or memory accesses. Routers are usually designed to be able to sustain a given packet processing rate, measured in packets per second [#fpps]_.
Packets per second versus bits per second
The performance of network nodes (either routers or switches) can be characterized by two key metrics :
the node's capacity measured in bits per second
the node's lookup performance measured in packets per second
The node's capacity in bits per second mainly depends on the physical interfaces that it uses and also on the capacity of the internal interconnection (bus, crossbar switch, ...) between the different interfaces inside the node. Many vendors, in particular for low-end devices will use the sum of the bandwidth of the nodes' interfaces as the node capacity in bits per second. Measurements do not always match this maximum theoretical capacity. A well designed network node will usually have a capacity in bits per second larger than the sum of its link capacities. Such nodes will usually reach this maximum capacity when forwarding large packets.
When a network node forwards small packets, its performance is usually limited by the number of lookup operations that it can perform every second. This lookup performance is measured in packets per second. The performance may depend on the length of the forwarded packets. The key performance factor is the number of minimal size packets that are forwarded by the node every second. This rate can lead to a capacity in bits per second which is much lower than the sum of the bandwidth of the node's links.
Let us now try to present a broad overview of the congestion problem in networks. We will assume that the network is composed of dedicated links having a fixed bandwidth [#fadjust]_. A network contains hosts that generate and receive packets and nodes (routers and switches) that forward packets. Assuming that each host is connected via a single link to the network, the largest demand is :math:`\sum{Access Links}`. In practice, this largest demand is never reached and the network will be engineered to sustain a much lower traffic demand. The difference between the worst-case traffic demand and the sustainable traffic demand can be large, up to several orders of magnitude. Fortunately, the hosts are not completely dumb and they can adapt their traffic demand to the current state of the network and the available bandwidth. For this, the hosts need to `sense` the current level of congestion and adjust their own traffic demand based on the estimated congestion. Network nodes can react in different ways to network congestion and hosts can sense the level of congestion in different ways.
Let us first explore which mechanisms can be used inside a network to control congestion and how these mechanisms can influence the behavior of the end hosts.
As explained earlier, one of the first manifestation of congestion on network nodes is the saturation of the network links that leads to a growth in the occupancy of the buffers of the node. This growth of the buffer occupancy implies that some packets will spend more time in the buffer and thus in the network. If hosts measure the network delays (e.g. by measuring the round-trip-time between the transmission of a packet and the return of the corresponding acknowledgment) they could start to sense congestion. On low bandwidth links, a growth in the buffer occupancy can lead to an increase of the delays which can be easily measured by the end hosts. On high bandwidth links, a few packets inside the buffer will cause a small variation in the delay which may not necessarily be larger that the natural fluctuations of the delay measurements.
If the buffer's occupancy continues to grow, it will overflow and packets will need to be discarded. Discarding packets during congestion is the second possible reaction of a network node to congestion. Before looking at how a node can discard packets, it is interesting to discuss qualitatively the impact of the buffer occupancy on the reliable delivery of data through a network. This is illustrated by the figure below, adapted from [Jain1990]_.

Loading…

No matching activity found.
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:178
String age
4 years ago
Source string age
4 years ago
Translation file
locale/pot/principles/sharing.pot, string 18