First, a little theory:
The concept of “Internet speed” does not exist; there is channel capacity, limited by a number of factors:
connection technology
tariff plan
the remoteness of a particular server you are working with
The bandwidth of an Internet channel is the maximum amount of data that can be received or transmitted per unit of time. The basic unit of throughput is bits per second. For larger values, larger units are used - kilobits per second / megabits per second / gigabits per second and so on. The bandwidth to each subscriber is determined both by the technology of its wired connection, it can be: ADSL or ADSL2+ up to 24 megabits per second, Fast Ethernet up to 100 megabits per second or Gigabit Ethernet up to 1000 megabits per second, and the selected tariff plan.
As written in the article about the features of using unlimited tariff plans, all these restrictions may depend on many factors, and if you are connected via Ethernet technology, which has a throughput of 100 megabits for both upload and download, it is not at all necessary that the server from which you will download anything, it will provide the same channel capacity.
The remoteness of the server is also of great importance. For example, you play online games, the online game server is located in Europe. The Infolink company does not have its own channels abroad, so when delays occur in the game, in most cases they are caused by the remoteness of the server, and losses on the channel, for example, between Helsinki and Amsterdam, which in no way applies to the Infolink company.
Carrying out bandwidth measurements to an external server
Find a point on the world map Noginsk and click on it.
Wait until the scan is completed.
Receipt speed will be the incoming current channel capacity. Transmission speed will be the outgoing current channel capacity.
The results, as a rule, are 5-10% less than real ones, because... do not take into account service traffic necessary for network operation. If you have followed all the recommendations, and the results obtained do not correspond to your connection technology and tariff plan, contact your provider's technical support service to resolve the problem.
Before measuring throughput, you should do the following to get the most objective results:
- to restart a computer;
- disable programs that use Internet resources, such as torrent clients, download managers, email clients, Internet radio and television, and others;
- suspend the operation of the antivirus, security center or firewall for the duration of the measurements.
- make sure that the proxy server is not set in the browser settings, because otherwise, measurements will be taken through it and the results may not even be close to reality.
-and the last thing is to connect directly to the provider’s cable without using a home router.
Internet bandwidth is primarily important for users, as it determines the speed of data transfer and comfortable work on the Internet.
It is estimated based on an analysis of the network’s ability to transmit information to one connected device. The data transfer rate depends on the selection of the optimal source, encoder and decoder for a given channel.
Separate the concepts of nominal and effective speed. Nominal determines the throughput when operating system and user applications, effective - only when the network is loaded with user programs.
To determine the speed of your Internet connection, special network tests are carried out. They allow you to measure the capabilities of the channel and determine the real speed, which, by the way, often does not correspond to what is stated in the tariff plan.
What does the speed test determine:
- incoming – an indicator of the operation of downloading data to your PC from the Internet;
- outgoing – the upload characteristic of transmitting information from your PC to the network;
- ping – the hourly period required to send a data packet from a PC to the provider’s server and back (determines the time the Internet page is opened);
- testing time – when the test was carried out, the result indicators can be compared with the current ones to check for changes.
The router's throughput depends on the type of cable (standard, fiber optic, etc.), the provider, and the current load on the network. As a result of the test, you do not always receive objective data, which rarely coincides with that stated in the tariff plan, but in any case, the throughput indicator should not deviate by more than 10%.
How to Determine Bandwidth
A capacity test is an elementary set of measures to verify compliance with the terms of the contract. It is also worth carrying out if you think that the Internet speed is lower than expected.
It is carried out using special online servers. Before starting the test, you should disable all download programs (torrent, mediaget, flashget, others). If Internet radio, email clients, Skype, ICQ and similar programs are running, they should also be terminated through the task manager. It is also recommended to close any antivirus programs that may be updating.
Testing instructions
Required minimum computer science.
As you know, data networks are designed to transmit information. Information is a special entity and is measured in specific units.
Since the data network is designed to transfer information primarily between computers, therefore the methods for measuring it are primarily computer-oriented.
In computer science, there is the concept of a bit - this is the minimum amount of information and it can have two states: yes - no, true - false, one - zero, etc.
A computer usually works not with individual bits, but with groups of them. A group containing 8 bits is called a byte. 8 bits = 1 byte.
Therefore, the amount of information is usually measured in numbers of bits or bytes.
To avoid confusion when abbreviating names, usually in most browsers and loader applications - the small Russian letter “b” or the small Latin “b” is designated “bit”, and the large letters “B” or “B” – “byte”.
Common mistakes when analyzing speed
A very common mistake when measuring speed is the incorrect interpretation of the data received from the downloader application, and you, seeing the value of 450 KBs, conclude that the speed is 10 times less than the stated tariff of 4096 kbs. But in the first case, the speed is measured in bytes, and if we multiply it by 8, we get 3600 kbit/s (taking into account the measurement error and service traffic, this is a completely acceptable result).
Another common mistake when measuring connection speed is that you look in the lower right corner of the monitor at the icon in the form of two monitors, and read the inscription “Connection speed 100 MB.” Then you remember that you have a tariff plan, for example, 512 Kbps, and since 512 is more than 100, the conclusion is drawn that you are being deceived, and you start calling technical support.
Let's find out what we're talking about! This icon and the inscription “Connection speed 100 MB” tell you the following information:
1) The modem is connected to the computer and there is a physical connection between them;
2) The modem and computer exchange information with each other at a speed of 100 Megabits per second (100 Mb/s).
3) In the case of FTTB connection, the role of the modem is performed by the home switch.
Currently, the speed of access to the Internet, even on the highest speed tariff plan, is lower than the speed of information exchange between the modem and PC. Therefore, this value has nothing to do with the speed of your Internet connection.
Bandwidth
So we will call the information transmission rate the amount of information, expressed in bits or bytes, transmitted per unit of time. Information transfer speed can be measured in bits per second - b/s, Kilobits per second - Kb/s or Megabits per second - Mb/s. Or in bytes per second - B/s, Kilobytes per second - KB/s, etc., respectively.
Another, very similar concept, which is often confused with the speed of information transmission, is channel throughput. It is measured in the same units as speed, but if the speed of information transmission shows how quickly information is transmitted from the source to the recipient, regardless of how and through what channels this information is transmitted, then the channel capacity shows how much information can be transmitted over specific data transmission channel per unit of time. Those. Bandwidth is the maximum possible data transfer rate for a particular channel.
In data transmission networks, one channel can simultaneously transmit information from many sources to many recipients and, depending on a number of factors, the information transmission speed for each specific source-receiver pair may be different, but the throughput for each channel is the same as usually constant.
The sum of all information transmission rates over a specific channel cannot be greater than the throughput of this channel!
No provider can guarantee the client in advance a given INFORMATION TRANSMISSION SPEED from/to any source of information from the network. The provider can only guarantee the client the CHANNEL CAPACITY.
Although the contracts and price lists of most providers indicate that the client is provided with such and such network access speed, in fact, this is not the speed, but the channel capacity.
And what channel? From a client in Krasnoyarsk to a server in the city of Uryupinsk or to the site www.windows.com?
No! The provider can guarantee the capacity of only those channels that belong to it. As a rule, this is a channel from the client to the provider's access channel to the global Internet, from the client to the provider's central node where its internal information resources are located, or from one client connection point to another. Also, to some extent, the provider is responsible for the capacity of its trunk channels to other network providers.
What determines the speed of information transfer?
Let's assume that you, as a client, measured the speed of information transfer from yourself (in Rostov-on-Don) to the server, say in Novosibirsk. Why did they “download” a large file from the server and record the time of its “downloading”. Then we divided the file size by the time and got the speed.
But for sure you will get a speed less than your declared “access speed” (that is, throughput). And your provider may not be to blame for this at all.
I'll try to explain why.
There may be three main reasons:
1) Overload of some communication channel between you and the Novosibirsk server. And there can be many channels there: from you to your provider, from the provider to its UpLink (“superior” provider), from the UpLink of your provider to the UpLink of the provider to which that same Novosibirsk server is connected (and in this there may be a rather long chain of channels belonging to different providers, including even foreign ones), as well as between the server and the provider to which it is connected. Moreover, the throughput of each of these channels may be different, and the “total” throughput of the entire channel will be no more than the throughput of the “slowest” of all “subchannels”.
2) Heavy load on the server itself (it simply slowly “gave” information to you), or restrictions on the speed of “upload” of data set by the server owner.
3) Low performance of your network equipment or heavy load on your computer with other tasks when you took measurements.
In addition, in this case you measured the “pure” speed of information transfer, so to speak, without any overhead costs. And there are quite a few of them: service information in the header of each IP packet, connection commands and installation of the information transfer process, resending of lost packets, etc. On average, these overhead costs are around 5-15%
Moreover, the higher the “access speed” you ordered from the provider, the more it may diverge from the information transfer speed measured in this way. Because in order to simply generate an information flow at a speed of more than 5 - 50 Mb/s, serious computing power is needed. From an ordinary personal computer with a budget network card, such measurements will have an accuracy of “plus or minus a big bast shoe”
How to measure speed correctly?
For some reason, many clients believe that every provider “sleeps and sees” how to deceive the client, how to give him an “access speed” that is lower than what he ordered.
This is wrong. Any serious provider tries to provide guaranteed throughput as accurately as possible, and not only because any client can measure it quite accurately and make a claim to the provider.
How to measure the throughput of a communication channel with a provider?
It is now fashionable among clients to measure “access speed” using various sites like speedtest.net. However, with the help of these sites you can only measure the speed of data transfer from you to this site, and not the bandwidth of your channel.
As I already wrote above, firstly, there are “two big differences”, secondly, the accuracy of such a measurement “leaves much to be desired” (for reasons stated in the previous section), thirdly, they can only show, so to speak, the “lower capacity limit, i.e. that the throughput is “not less” than what you intended.
The most reliable way to measure the true throughput of your channel is as follows.
First of all, you need to have some kind of program that can calculate the amount of transmitted/received information directly on the interface of your computer - such as TMeter, DUMeter, etc.
After launching such a program, you need to “download” your channel as much as possible in any way, for example, start “downloading” simultaneously several fairly large files from different FTP servers (and the more, the better). Or another way is to launch the popular application today - Torrent, typing as many downloads into it as possible, and evaluate the overall download speed. Then you will be able to accurately determine exactly the bandwidth of your channel to the provider, because more information than the provider allows you “simply will not get through” to your computer.
A little about ADSL
There are also cases when the provider cannot provide the channel bandwidth between you and its network, in accordance with the tariff plan you have chosen. This most often happens in cases of ADSL connections. If you have studied the operation of DSL access technologies, then you should know that the throughput of this channel largely depends on the length of the subscriber line, the thickness of the core, the quality of the cable laying and its age. So, in some cases, the provider does not have the technical ability to provide you with an ADSL connection with the maximum bandwidth allowed with this technology of 25 Mbps. Therefore, for most lines the norm is a value of 6-8 Mbps.
In today's IP networks, with the emergence of many new network applications, it becomes increasingly difficult to estimate the required bandwidth: typically, you need to know what applications you plan to use, what data protocols they use, and how they will communicate
Ilya Nazarov
System engineer at INTELCOM Line
After assessing the required throughput on each section of the IP network, it is necessary to decide on the choice of OSI network and link layer technologies. In accordance with the selected technologies, the most suitable models of network equipment are determined. This question is also difficult, since throughput directly depends on hardware performance, and performance, in turn, depends on the hardware and software architecture. Let's take a closer look at the criteria and methods for assessing the capacity of channels and equipment in IP networks.
Bandwidth Evaluation Criteria
Since the emergence of teletraffic theory, many methods have been developed for calculating channel capacity. However, unlike calculation methods applied to circuit-switched networks, calculating the required throughput in packet networks is quite complex and is unlikely to provide accurate results. First of all, this is due to a huge number of factors (especially those inherent in modern multiservice networks), which are quite difficult to predict. In IP networks, a common infrastructure is typically used by many applications, each of which may use its own different traffic pattern. Moreover, within one session, traffic transmitted in the forward direction may differ from traffic transmitted in the opposite direction. In addition to this, calculations are complicated by the fact that the speed of traffic between individual network nodes can change. Therefore, in most cases when building networks, the capacity assessment is actually determined by the general recommendations of manufacturers, statistical studies and the experience of other organizations.
To more or less accurately determine how much bandwidth is required for the network being designed, you must first know what applications will be used. Next, for each application, you should analyze how data will be transferred during the selected periods of time, and what protocols are used for this.
For a simple example, consider applications on a small corporate network.
Example of bandwidth calculation
Let's assume there are 300 work computers and the same number of IP phones on the network. It is planned to use the following services: email, IP telephony, video surveillance (Fig. 1). For video surveillance, 20 cameras are used, from which video streams are transmitted to the server. Let's try to estimate what maximum bandwidth is required for all services on the channels between the network core switches and at the junctions with each of the servers.
It should be noted right away that all calculations must be carried out for the time of greatest network activity of users (in teletraffic theory - peak hours), since usually during such periods network performance is most important and delays and failures in application operation associated with a lack of bandwidth occur. , are unacceptable. In organizations, the greatest load on the network may occur, for example, at the end of the reporting period or during a seasonal influx of customers, when the largest number of telephone calls are made and the majority of email messages are sent.
Email
Returning to our example, consider an email service. It uses protocols that run on top of TCP, meaning the data transfer rate is constantly adjusted to take up all the available bandwidth. Thus, we will start from the maximum delay value for sending a message - let’s say 1 second will be enough to make the user comfortable. Next, you need to estimate the average size of the message sent. Let's assume that during peak activity, email messages will often contain various attachments (copies of invoices, reports, etc.), so for our example we'll take the average message size to be 500 KB. Finally, the last parameter we need to select is the maximum number of employees who can simultaneously send messages. Let's say that during emergency times, half of the employees simultaneously press the "Send" button in the email client. The required maximum throughput for email traffic would then be (500 kB x 150 hosts)/1 s = 75,000 kB/s or 600 Mbps. From here we can immediately conclude that to connect the mail server to the network it is necessary to use a Gigabit Ethernet channel. At the core of the network, this value will be one of the terms that makes up the total required throughput.
Telephony and video surveillance
Other applications - telephony and video surveillance - are similar in their stream transmission structure: both types of traffic are transmitted using the UDP protocol and have a more or less fixed transmission rate. The main differences are that in telephony the streams are bidirectional and limited by the time of the call, while in video surveillance the streams are transmitted in one direction and, as a rule, are continuous.
To estimate the required throughput for telephony traffic, assume that during peak activity the number of simultaneous connections passing through the gateway can reach 100. When using the G.711 codec on Ethernet networks, the speed of one stream, taking into account headers and service packets, is approximately 100 kbit/s. With. Thus, during periods of greatest user activity, the required bandwidth in the network core will be 10 Mbit/s.
Video surveillance traffic is calculated quite simply and accurately. Let’s say that in our case, video cameras transmit streams of 4 Mbit/s each. The required bandwidth will be equal to the sum of the speeds of all video streams: 4 Mbit/s x 20 cameras = 80 Mbit/s.
All that remains is to add up the resulting peak values for each of the network services: 600 + 10 + 80 = 690 Mbit/s. This will be the required bandwidth in the network core. The design should also include the possibility of scaling so that communication channels can serve the traffic of a growing network for as long as possible. In our example, it will be enough to use Gigabit Ethernet to meet the requirements of the services and at the same time be able to seamlessly develop the network by connecting more nodes
Of course, the example given is far from being a standard one - each case must be considered separately. In reality, the network topology can be much more complex (Fig. 2), and capacity assessment must be made for each section of the network.
It should be taken into account that VoIP traffic (IP telephony) is distributed not only from phones to the server, but also between phones directly. In addition, network activity may vary in different departments of the organization: the technical support service makes more phone calls, the project department uses e-mail more actively than others, the engineering department consumes more Internet traffic than others, etc. As a result, some parts of the network may require more bandwidth than others.
Usable and full throughput
In our example, when calculating the IP telephony flow rate, we took into account the codec used and the size of the packet header. This is an important detail to keep in mind. Depending on the encoding method (codecs used), the amount of data transmitted in each packet, and the link-layer protocols used, the total throughput of the stream is formed. It is the total throughput that must be taken into account when estimating the required network throughput. This is most relevant for IP telephony and other applications that use real-time transmission of low-speed streams, in which the size of the packet headers is a significant part of the size of the entire packet. For clarity, let's compare two VoIP streams (see table). These streams use the same compression, but different payload sizes (actually, the digital audio stream) and different link layer protocols.
The data transfer rate in its pure form, without taking into account network protocol headers (in our case, a digital audio stream), is useful bandwidth. As you can see from the table, with the same useful throughput of streams, their total throughput can vary greatly. Thus, when calculating the required network capacity for telephone calls during peak loads, especially for telecom operators, the choice of channel protocols and flow parameters plays a significant role.
Equipment selection
The choice of link-layer protocols is usually not a problem (today the question more often arises of how much bandwidth an Ethernet channel should have), but choosing the right equipment can cause difficulties even for an experienced engineer.
The development of network technologies, along with the growing demands of applications for network bandwidth, forces network equipment manufacturers to develop ever new software and hardware architectures. Often, from a single manufacturer there are seemingly similar equipment models, but designed to solve different network problems. Take, for example, Ethernet switches: most manufacturers, along with conventional switches used in enterprises, have switches for building data storage networks, organizing operator services, etc. Models of the same price category differ in their architecture, “tailored” for specific tasks.
In addition to overall performance, the choice of equipment should also be based on supported technologies. Depending on the type of hardware, a certain set of functions and types of traffic can be processed at the hardware level without using CPU and memory resources. At the same time, traffic from other applications will be processed at the software level, which greatly reduces overall performance and, as a result, maximum throughput. For example, multi-layer switches, thanks to their complex hardware architecture, are capable of transmitting IP packets without reducing performance when all ports are at maximum load. Moreover, if we want to use more complex encapsulation (GRE, MPLS), then such switches (at least inexpensive models) are unlikely to suit us, since their architecture does not support the corresponding protocols, and at best such encapsulation will occur at the expense of the central processor low productivity. Therefore, to solve such problems, we can consider, for example, routers whose architecture is based on a high-performance central processor and depends to a greater extent on software rather than hardware implementation. In this case, at the expense of maximum throughput, we get a huge range of supported protocols and technologies that are not supported by switches in the same price category.
Overall Equipment Performance
In the documentation for their equipment, manufacturers often indicate two maximum throughput values: one expressed in packets per second, the other in bits per second. This is due to the fact that most of the performance of network equipment is spent, as a rule, on processing packet headers. Roughly speaking, the equipment must receive the packet, find a suitable switching path for it, generate a new header (if necessary) and transmit it further. Obviously, in this case it is not the volume of data transmitted per unit of time that plays a role, but the number of packets.
If you compare two streams transmitted at the same speed but with different packet sizes, then the stream with a smaller packet size will require more performance to transmit. This fact should be taken into account if, for example, a large number of IP telephony streams are supposed to be used on the network - the maximum throughput in bits per second here will be much less than declared.
It is clear that with mixed traffic, and even taking into account additional services (NAT, VPN), as happens in the vast majority of cases, it is very difficult to calculate the load on equipment resources. Often, equipment manufacturers or their partners load test different models under different conditions and publish the results on the Internet in the form of comparison tables. Familiarization with these results greatly simplifies the task of choosing the appropriate model.
Pitfalls of modular equipment
If the selected network equipment is modular, then in addition to the flexible configuration and scalability promised by the manufacturer, you can get many pitfalls.
When choosing modules, you should carefully read their description or consult the manufacturer. It is not enough to be guided only by the type of interfaces and their number - you also need to become familiar with the architecture of the module itself. For similar modules, it is not uncommon that when transmitting traffic, some are able to process packets autonomously, while others simply forward packets to the central processing module for further processing (accordingly, for externally identical modules, the price for them can differ several times). In the first case, the overall performance of the equipment and, as a consequence, its maximum throughput are higher than in the second, since the central processor shifts part of its work to the processors of the modules.
In addition, modular equipment often has a blocking architecture (when the maximum throughput is lower than the total speed of all ports). This is due to the limited capacity of the internal bus through which the modules exchange traffic with each other. For example, if a modular switch has a 20 Gbps internal bus, its 48-port Gigabit Ethernet line card can only use 20 ports when fully loaded. You should also keep such details in mind and carefully read the documentation when choosing equipment.
When designing IP networks, bandwidth is a key parameter that will determine the architecture of the network as a whole. For a more accurate assessment of throughput, you can follow the following recommendations:
- Study the applications that you plan to use on the network, the technologies they use, and the volume of transmitted traffic. Use the advice of developers and the experience of colleagues to take into account all the nuances of these applications when building networks.
- Dive deep into the network protocols and technologies used by these applications.
- Read the documentation carefully when choosing equipment. To have some stock of ready-made solutions, check out the product lines of different manufacturers.
As a result, with the right choice of technologies and equipment, you can be sure that the network will fully satisfy the requirements of all applications and, being sufficiently flexible and scalable, will last for a long time.
Ilya NazarovSystem engineer at INTELCOM Line
After assessing the required throughput on each section of the IP network, it is necessary to decide on the choice of OSI network and link layer technologies. In accordance with the selected technologies, the most suitable models of network equipment are determined. This question is also difficult, since throughput directly depends on hardware performance, and performance, in turn, depends on the hardware and software architecture. Let's take a closer look at the criteria and methods for assessing the capacity of channels and equipment in IP networks.
Bandwidth Evaluation Criteria
Since the emergence of teletraffic theory, many methods have been developed for calculating channel capacity. However, unlike calculation methods applied to circuit-switched networks, calculating the required throughput in packet networks is quite complex and is unlikely to provide accurate results. First of all, this is due to a huge number of factors (especially those inherent in modern multiservice networks), which are quite difficult to predict. In IP networks, a common infrastructure is typically used by many applications, each of which may use its own different traffic pattern. Moreover, within one session, traffic transmitted in the forward direction may differ from traffic transmitted in the opposite direction. In addition to this, calculations are complicated by the fact that the speed of traffic between individual network nodes can change. Therefore, in most cases when building networks, the capacity assessment is actually determined by the general recommendations of manufacturers, statistical studies and the experience of other organizations.
To more or less accurately determine how much bandwidth is required for the network being designed, you must first know what applications will be used. Next, for each application, you should analyze how data will be transferred during the selected periods of time, and what protocols are used for this.
For a simple example, consider applications on a small corporate network.
Example of bandwidth calculation
Let's assume there are 300 work computers and the same number of IP phones on the network. It is planned to use the following services: email, IP telephony, video surveillance (Fig. 1). For video surveillance, 20 cameras are used, from which video streams are transmitted to the server. Let's try to estimate what maximum bandwidth is required for all services on the channels between the network core switches and at the junctions with each of the servers.
It should be noted right away that all calculations must be carried out for the time of greatest network activity of users (in teletraffic theory - peak hours), since usually during such periods network performance is most important and delays and failures in application operation associated with a lack of bandwidth occur. , are unacceptable. In organizations, the greatest load on the network may occur, for example, at the end of the reporting period or during a seasonal influx of customers, when the largest number of telephone calls are made and the majority of email messages are sent.
Email
Returning to our example, consider an email service. It uses protocols that run on top of TCP, meaning the data transfer rate is constantly adjusted to take up all the available bandwidth. Thus, we will start from the maximum delay value for sending a message - let’s say 1 second will be enough to make the user comfortable. Next, you need to estimate the average size of the message sent. Let's assume that during peak activity, email messages will often contain various attachments (copies of invoices, reports, etc.), so for our example we'll take the average message size to be 500 KB. Finally, the last parameter we need to select is the maximum number of employees who can simultaneously send messages. Let's say that during emergency times, half of the employees simultaneously press the "Send" button in the email client. The required maximum throughput for email traffic would then be (500 kB x 150 hosts)/1 s = 75,000 kB/s or 600 Mbps. From here we can immediately conclude that to connect the mail server to the network it is necessary to use a Gigabit Ethernet channel. At the core of the network, this value will be one of the terms that makes up the total required throughput.
Telephony and video surveillance
Other applications - telephony and video surveillance - are similar in their stream transmission structure: both types of traffic are transmitted using the UDP protocol and have a more or less fixed transmission rate. The main differences are that in telephony the streams are bidirectional and limited by the time of the call, while in video surveillance the streams are transmitted in one direction and, as a rule, are continuous.
To estimate the required throughput for telephony traffic, assume that during peak activity the number of simultaneous connections passing through the gateway can reach 100. When using the G.711 codec on Ethernet networks, the speed of one stream, taking into account headers and service packets, is approximately 100 kbit/s. With. Thus, during periods of greatest user activity, the required bandwidth in the network core will be 10 Mbit/s.
Video surveillance traffic is calculated quite simply and accurately. Let’s say that in our case, video cameras transmit streams of 4 Mbit/s each. The required bandwidth will be equal to the sum of the speeds of all video streams: 4 Mbit/s x 20 cameras = 80 Mbit/s.
All that remains is to add up the resulting peak values for each of the network services: 600 + 10 + 80 = 690 Mbit/s. This will be the required bandwidth in the network core. The design should also include the possibility of scaling so that communication channels can serve the traffic of a growing network for as long as possible. In our example, it will be enough to use Gigabit Ethernet to meet the requirements of the services and at the same time be able to seamlessly develop the network by connecting more nodes
Of course, the example given is far from being a standard one - each case must be considered separately. In reality, the network topology can be much more complex (Fig. 2), and capacity assessment must be made for each section of the network.
It should be taken into account that VoIP traffic (IP telephony) is distributed not only from phones to the server, but also between phones directly. In addition, network activity may vary in different departments of the organization: the technical support service makes more phone calls, the project department uses e-mail more actively than others, the engineering department consumes more Internet traffic than others, etc. As a result, some parts of the network may require more bandwidth than others.
Usable and full throughput
In our example, when calculating the IP telephony flow rate, we took into account the codec used and the size of the packet header. This is an important detail to keep in mind. Depending on the encoding method (codecs used), the amount of data transmitted in each packet, and the link-layer protocols used, the total throughput of the stream is formed. It is the total throughput that must be taken into account when estimating the required network throughput. This is most relevant for IP telephony and other applications that use real-time transmission of low-speed streams, in which the size of the packet headers is a significant part of the size of the entire packet. For clarity, let's compare two VoIP streams (see table). These streams use the same compression, but different payload sizes (actually, the digital audio stream) and different link layer protocols.
The data transfer rate in its pure form, without taking into account network protocol headers (in our case, a digital audio stream), is useful bandwidth. As you can see from the table, with the same useful throughput of streams, their total throughput can vary greatly. Thus, when calculating the required network capacity for telephone calls during peak loads, especially for telecom operators, the choice of channel protocols and flow parameters plays a significant role.
Equipment selection
The choice of link-layer protocols is usually not a problem (today the question more often arises of how much bandwidth an Ethernet channel should have), but choosing the right equipment can cause difficulties even for an experienced engineer.
The development of network technologies, along with the growing demands of applications for network bandwidth, forces network equipment manufacturers to develop ever new software and hardware architectures. Often, from a single manufacturer there are seemingly similar equipment models, but designed to solve different network problems. Take, for example, Ethernet switches: most manufacturers, along with conventional switches used in enterprises, have switches for building data storage networks, organizing operator services, etc. Models of the same price category differ in their architecture, “tailored” for specific tasks.
In addition to overall performance, the choice of equipment should also be based on supported technologies. Depending on the type of hardware, a certain set of functions and types of traffic can be processed at the hardware level without using CPU and memory resources. At the same time, traffic from other applications will be processed at the software level, which greatly reduces overall performance and, as a result, maximum throughput. For example, multi-layer switches, thanks to their complex hardware architecture, are capable of transmitting IP packets without reducing performance when all ports are at maximum load. Moreover, if we want to use more complex encapsulation (GRE, MPLS), then such switches (at least inexpensive models) are unlikely to suit us, since their architecture does not support the corresponding protocols, and at best such encapsulation will occur at the expense of the central processor low productivity. Therefore, to solve such problems, we can consider, for example, routers whose architecture is based on a high-performance central processor and depends to a greater extent on software rather than hardware implementation. In this case, at the expense of maximum throughput, we get a huge range of supported protocols and technologies that are not supported by switches in the same price category.
Overall Equipment Performance
In the documentation for their equipment, manufacturers often indicate two maximum throughput values: one expressed in packets per second, the other in bits per second. This is due to the fact that most of the performance of network equipment is spent, as a rule, on processing packet headers. Roughly speaking, the equipment must receive the packet, find a suitable switching path for it, generate a new header (if necessary) and transmit it further. Obviously, in this case it is not the volume of data transmitted per unit of time that plays a role, but the number of packets.
If you compare two streams transmitted at the same speed but with different packet sizes, then the stream with a smaller packet size will require more performance to transmit. This fact should be taken into account if, for example, a large number of IP telephony streams are supposed to be used on the network - the maximum throughput in bits per second here will be much less than declared.
It is clear that with mixed traffic, and even taking into account additional services (NAT, VPN), as happens in the vast majority of cases, it is very difficult to calculate the load on equipment resources. Often, equipment manufacturers or their partners load test different models under different conditions and publish the results on the Internet in the form of comparison tables. Familiarization with these results greatly simplifies the task of choosing the appropriate model.
Pitfalls of modular equipment
If the selected network equipment is modular, then in addition to the flexible configuration and scalability promised by the manufacturer, you can get many pitfalls.
When choosing modules, you should carefully read their description or consult the manufacturer. It is not enough to be guided only by the type of interfaces and their number - you also need to become familiar with the architecture of the module itself. For similar modules, it is not uncommon that when transmitting traffic, some are able to process packets autonomously, while others simply forward packets to the central processing module for further processing (accordingly, for externally identical modules, the price for them can differ several times). In the first case, the overall performance of the equipment and, as a consequence, its maximum throughput are higher than in the second, since the central processor shifts part of its work to the processors of the modules.
In addition, modular equipment often has a blocking architecture (when the maximum throughput is lower than the total speed of all ports). This is due to the limited capacity of the internal bus through which the modules exchange traffic with each other. For example, if a modular switch has a 20 Gbps internal bus, its 48-port Gigabit Ethernet line card can only use 20 ports when fully loaded. You should also keep such details in mind and carefully read the documentation when choosing equipment.
When designing IP networks, bandwidth is a key parameter that will determine the architecture of the network as a whole. For a more accurate assessment of throughput, you can follow the following recommendations:
- Study the applications that you plan to use on the network, the technologies they use, and the volume of transmitted traffic. Use the advice of developers and the experience of colleagues to take into account all the nuances of these applications when building networks.
- Dive deep into the network protocols and technologies used by these applications.
- Read the documentation carefully when choosing equipment. To have some stock of ready-made solutions, check out the product lines of different manufacturers.
As a result, with the right choice of technologies and equipment, you can be sure that the network will fully satisfy the requirements of all applications and, being sufficiently flexible and scalable, will last for a long time.
Loading...
