Before what to engage in consideration of parameters multimedia traffic, will talk about what exactly multimedia applications more frequent than all meet in modern networks. It will allow us more precisely to define an object our researches.

Let's divide all of applications, requiring a transmission to multimedia information in real time, on two large class: interactive and uninteractive. Appendixes of the first groups are IP-telephony and conferences. Their distinctive a feature is in that participants, as a rule, two or a few. Every participant passes to information all other. In it turn, each of gettings it participants reacts on it (for example, answers on a remark), and information is passed from answering all of other. Thus, the system turns out with by a feed-back. In addition, desirably, that feed-back acted in good time, otherwise a talk will be an uncomfort. Other a situation is observed in work of uninteractive applications. To it the so-called 'systems belong a group audio and video to on to the query', examples of which are the wireless internetstations and digital telecasting. As a rule, the similar systems are had in the basis client-server technology: a client unites with a server and sends to him a request for delivery of multimedia stream. Server gives out a stream which a client can begin to decode' at once after a receipt, not waiting for completion of stream. Distinctive a feature of the uninteractive systems is in that the same stream can look over at once very much a lot of users, because feed-back from each of them, intended other to the participants, not required.

Programs of both types, both interactive and uninteractive, it is important and it is needed to probe, but uninteractive probing appendixes is simpler, because their use carries mass character. For example, work is devoted research multimedia traffic which was consumed by the clients of network Washingtonian university (THE USA). Basic results researches such: for one week 4786 clients appealed to to to 23738 different objects on protocol of RTSP, it was here 56 Gb of information are carried. In addition, were got important additional results: loading, created on a network by the combined stream, was cyclic with a period in one days, at it a peak of carrying capacity was on a workings clock workings days of week and 2,8 made Mbit/sek. Most sessions lasted men'she 10 minutes, and volume carried by them information made men'she 1 MByte. At the same time, just 3% sessions passed the half of the combined volume of information almost. The scale of this research allows in a sufficient measure to extrapolate his results for description of multimedia streams in networks.

Of these results it is not enough yet, to reflect structure of multimedia traffic. We will appeal to work, in which was probe the multimedia streams of format of Realaudio from the servers of company 'Broadcast.com' to the clients (a measuring equipment was on a technical ground to the company Broadcast.com). Transport was utillized in this case protocol of PNA, but not RTSP, as in work. Formats indicated higher encoding and communications of data are the closed development companies 'Real Networks', but their wide practical prevalence induces to their study. During research it turned out that greater part of sessions (70-80%) had utillized two stream: one, on protocol of UDP, for communication of data, second, to on to protocol of TCP, for flow control. Remaining 20-30% sessions utillize an only one stream, protocol of TCP, and for information, and for a management. A management stream serves for creation of reverse connections with a server: on him confirmations of accepted are sent packages, and also command from an user, for example, on halt or stopping of stream. Volume of information, passed back to a server, was very insignificant, his attitude to toward toward a client made the volume of information about 1:28..1:50.

Was found that the size of packages depends on to bit speed of audiostream. If as multimedia content stereozvuk comes forward, bit speed is utillized 16 and 20 Kbit/s, and sizes of packages are 293 and 495 byte. In case of transmission of vocal stream, bit speed prevails 6,5 Kbit/s with the size of packages 244 bytes.

Analysis showed that packages were sent not through the equal intervals of time, but by packs (bursts). Thus observed on the average for 6 packages in a pack, followed whereupon pause on 1,8 with or (rarer) multiple by it value. It engineer a deep practical ground has a decision. Sometimes it is considered that, if a multimedia traffic is passed with by permanent bit speed, in practice it is necessary to send packages through the equal intervals of time. Actually, permanent bit speed is indeed observed, but on large time domains (for example, with usredneniem for a minute). On the dispatch of every package a quantum is required an appendix-server to time from the planner of tasks of the operating system. If to send packages through the equal intervals of time, required frequent context switching by the planner of tasks, that is attended with high overhead costs. Therefore more advantageous appears in time to send away at once a few slice selected a planner packages. In case if the general load of server grows, to the planner it is not succeeded fluently to distribute time between tasks, that affects work of server of Realaudio: unlike ordinary mode, when after a pack the short ensues from a few packages interval, a server begins to increase duration of interval between packs, and more packages turn out in a pack. Bit speed of stream at usrednenii for a minute will remain former, but in small time scale factors a dispatch of packages already will not be smooth. This example shows that for effective work multimedia server required also and effective operating system. Processes of server on the nature need planning in the real scale of time. Proper expansions for the planners of tasks described the standard of POSIX 1003.1b. Some taking about this standard and his support it is possible in OS of Linux and SUNOS to draw from work.

Work the model of RealAudio-trafic was built, after what it was tested by the system of imitation designs of ns-2. Verification showed a good coincidence to the built model with the results of the practical measurings. There are of interest models of mul'tipleksirovannykh multimedia streams, generated a few sources at once. Such models allow to get analytical expression for distribution of probabilities of the state of buffer. In particular, described puassonovskiy process (MMPP-процесс) receipts of packages are in a mul'tipleksirovannom stream. For this purpose the discrete chain of Markova is examined with an eventual number states of M. Accepted also, that if the system is in state of m, where m=1..M, on the entrances of attendant device a stream acts with intensity of m . The simplest practically interesnym a case will be M=2, that two states systems: for example, large and small loading, created by a mul'tipleksirovannym stream and proper to two different to the values of intensity: 1 and 2 . However drafting of model with requires the decision of the system of nonlinear the help of MMPP-process equalizations for finding of parameters of markovskoy chain. In such case it is possible to take advantage of the model liquid stream (fluid-flow). This model supposes that each an active presently source generates one unit information in one time unit, and attendant device gets on an entrance information at a speed of C of units of information in unit to time from C of active presently sources. Then for descriptions of stream from M of sources part of which can be active at any moment to time, a markovskaya chain will be required with M by the states. After drafting and decision of the system of linear differential equalizations, it is possible to find distributing probabilities of the states of buffer of Fi x is probability that in set mode at activity of I sources in a buffer it will be to be no more than x units of information.

Perception by the man of multimedia it will be information the best only at transmissions multimedia stream on a network without distortions. At delays and losses of packages perception will suffer. Nevertheless, if deviations of descriptions of stream from nominal are in some limits, quality of stream will be perceived as possible. If at the set level of the multimedia loading on network quality of reproducing of information remains possible, modernization of network will be economic groundless. To to to finding out of limits of rejections of parameters of stream, not influencing by substantial appearance on perception of information by a man, we now and passed.

Research of computer network on a fitness to the transmission of multimedia information

Review of existent birth-certificates is in Ip-setyakh

Esche one important requirement to measurings consists in that, that the sizes of IP-пакетов in a testing stream must be less than minimum value MTU (Maximum Transfer Unit) all interfaces of every device of network layer along a route, to avoid fragmentation of IP-пакетов and related to it distortions of results of measurings. To know MTU along a route it is possible by the utility of tracepath, included in the package of iputils in Gnu/linux.

Method of research of computer network on fitness to the transmission of multimedia traffic

Was shown before, for the transmission of multimedia traffic the most essential parameters of network it is been delay, brought in a network, and level of losses of packages, because they in the first turn influence on perception of stream an user. It was therefore it is decided at creation of method of research of computer network to measure these parameters exactly.

Sufficient for practical aims exactness of results a method must possess the minimum cost of raising experiment and analysis of results. Therefore during work it was freely expandable software is applied: OS of Gnu/linux and programs, included in its composition. One of hosts, participating in measuring, was in the network of Spbgpu, and second - in networks of the Petrodvorcovogo telecommunication center Spbgu. The kernel of programmatic complex is the program MGEN [8] - generator and receiver of test traffic, which was created by the specialists of military department of the USA specially for an analysis computer networks. Program MGEN, set on one computer of network, generates a test traffic, and the same programa on other computer accepts him and writes down in the file of magazine information about the accepted packages, which plugs in itself: time dispatches of package on the local clock of sender, sequence number sent package (two these the values are sent in each package), and also time of his reception on the local clock of receiver. Work of MGEN is managed a script file. Chart of experiment represented on a fig. 1.

for the host of 'Master' in the environment of OS Gnu/linux on the basis of command of at (1) an infrastructure which through time domains was created, having the exponential distributing, started a script test_mgen_two_way. This script executed the followings tasks:

1. Read measuring system configuration from a file

2. Started the script of listen_mgen on the host of 'Slave'

3. Checked, whether there was he started on the host of 'Master' or on a host 30 'Slave'. If was started on 'Master', started it copy of test_mgen_two_way on the host of 'Slave'

4. MGEN started with the scenario of send.mgn for a dispatch test traffic to the remote host

5. After completion of dispatch of test traffic appointed time the next start through a casual interval to time, selectable in obedience to exponential to distributing

skript listen_mgen executed the followings tasks:

1. Read measuring system configuration from a file

2. MGEN started with the scenario of receive.mgn for a reception test traffic from a remote host

3. After completion of reception of test traffic sent a file with by the results of measurings on the address of e-mail starting him user, whereupon wrote down a file in catalogue for storage

Automatickly start of the programs on a remote computer took a place with the use of protocol of SSH and his opened realization of OPENSSH [10]. As a result of joint work described system the host of 'Master' was passed by a test traffic to the host 'Slave' and vice versa. Thus, both hosts are perfect ravnopravny at the dispatch of traffic, and one of computers is adopted 'Master' because he initiates measuring. After completion measurings the file of magazine which appeared on each of hosts referred by e-mail (aggregate of two files of magazine, one by one from every host, in future will be named 'Track experiment'). Further the results of measurings were analysed and drawn a conclusion about quality of that segment of computer network, which connected two hosts.

In time of measuring a traffic was passed in both sides at the same time, that imitates the typical consisting of case conducting of a few multimedia conferences in a network between by hosts: a stream from every host to his neighbour is analogical to the mul'tipleksirovannomu stream a few simultaneous conferences. If it is assumed that a network will be utillized mainly for uninteractive applications, such, as Realaudio, when almost all of traffic is passed only aside client, it is possible to change testing system configuration. At it every host becomes conducting and regardless of neighbour chooses an interval between measurings coming from exponential distributing.

In composition of MGEN enters program of trpr, which allows to extract from the files of magazine of every host information about a delay transmissions of every package from an end in an end, about a time domain between the receipts of nearby packages and about the stake of the lost packages during measuring. Nevertheless, a clock of two hosts was not synchronized with each other, therefore measuring results one-sided delay were useless for direct application. In tabl. 2 terminology is considered for the specification of parameters of clock. Implied, that a 'exact' clock is, with testimonies which we will compare testimonies of examinee clock.

Chasy of two hosts, utillized as a result of experiment, possess failings: at first, they not sinkhronizirovanny ( I am possess small exactness: the module of change is not near to the zero), vo- second, they go along with different speed: a frequency change is not equal to the zero. It is possible to conclude in relation to exactness of clock, that unnecessarily to synchronize the clock of both hosts with some by 'exact' hours. It is instead possible to synchronize a clock one host with a clock other. It would enable to determine delay of transmission of package in one side. However in the case of our measurings was inaccessible even it. Synchronization of clock to on does not fit protocol of NTP [13], because he is intended for synchronizations of time with exactness of a few ten milliseconds during time domains in a few days. In our case synchronization was required with exactness about 1 ms on a small segment time, determined by duration measurings. In addition, use of protocol of NTP even on one of hosts can result in unexpected effects in time measurings, because the timer of host, utillizing NTP, is moved in measuring motion forward or back [24]. By the cardinal decision of synchronization of testimonies of two clock there is the system of the global positioning of GPS, created by the military department of the USA. Organization of RIPE NCC (European network co-ordinating center) created programmno-apparatnyy complex for measuring of parameters of IP-сетей [22]. Measuring device the station is a computer, connected with a receiver systems of GPS. During work a computer timer gives out testimonies, synchronized with the system of GPS. Exactness of testimonies of GPS- receiver so great, that exactness of temporal marks in begins the measuring system to be limited to permission timer and makes about 1 ms, that fully sufficiently for practical aims. Developed programmno-apparatnyy a complex stands €2500 and is for a sale actually at cost. Organizations, solicitous of to participate in testing connections between by the network and rest of Internet, can purchase such device and to set him on the technical ground. After options a device works without intervention from an operator and passes collected statistics for an analysis in RIPE NCC, doing its accessible for all of participants of measurings.

To circumstance, that a frequency change between a clock is not equal to the zero, requires an obligatory account. This failing is expressed clock in that they go along with other speed (let, for to definiteness, they move a bit faster), what their pair. To on a frequency change is small an absolute value, for example, in motion measurings there was a value of change of testimonies 1 ms for each 5,5 with. But for 180 seconds, which are lasted by measuring, testimonies of clock moved in relation to each other on 32 msecs To this it is impossible a size to scorn at an analysis. Nevertheless, on short intervals to time, comparable with duration of measuring, frequency change often is permanent a size (during measurings of rejections from this rule it was not), then a change of testimonies of clock will be linear, and he can be deleted after the receipt of results measurings. The delete of frequency change, thus, removes methodical error of measurings.

Was mentioned, that nesinkhronizirovannost' clock did direct use of measuring results impossible. High-quality a picture looks like the following. If change between the testimonies of clock makes, it admits, 10 with, measured in one side a delay appeared approximately 10 is equal with, and in other side - approximately -10 with. Real value of one-sided delays from these information, getting is impossible. Nevertheless, it is possible to get the value of delay in both sides.

Ofthen under RTT (Round-trip Time, 'time of complete turn') imply time which is required a package, to attain addressee and to return back. Measuring of RTT is often made with by the help of the program of ping and similar to it. Such programs message protocol of ICMP [20] 'Echo request', it arrives at an addressee, the kernel of the operating system of which generates return report of ICMP 'Echo reply', which leaves back. Sender, getting an answer for the own report, calculates, how many time passed from the moment of dispatch, and got value and RTT is named. The indicated method possesses failing: ICMP-пакеты not necessarily will be served on intermediate and eventual knots the same as and packages with by information. Often routers limit a carrying capacity, given the stream of ICMP-пакетов, from considering to safety [30]. In addition, IP-стек of routers and receiver can work with ICMP-пакетами as with by less priority and to generate and move forward them only in case that on service there are not packages with by information [24]. It results in that value RTT, looked after at such testing, reflect not time of complete turn of packages with by information, and time of complete turn ICMP-пакетов, that is quite not required a researcher. Exists improved variant of this method, when on a passing host the generator of packages is set with by information, but not on protocol of ICMP, and, for example, on UDP. There is a receiver on other host. At receipt of package a receiver immediately sends him back. A sender gets an answer and calculates a value RTT the same as w I l l o w s previous case. The feature of chart consists of that the not packages of ICMP, but packages of UDP, are utillized, and looked after a value RTT reflects terms in which passed better pol'zovatl'skie information. In addition, 'mirror reflection' of package does the not kernel of OS of receiver, but appendix on a receiver, therefore the looked after value RTT is yet a bit nearer to that which would test the real flows of data.

It was planned this work to create the method of estimation of RTT not one of the methods indicated higher, and by addition of values two one-sided delays. For this purpose it was necessary to act so: to delete frequency change from track of experiment, to calculate a median looked after values of one-sided delay from the host of 'Master' to to to the host of 'Slave', after to calculate a median analogical appearance values of one-sided delay from 'Slave' to 'Master'. After it to lay down the got medians and, in obedience to a method essence of which expounded before, to get as a result a value RTT for given experiment. Dignity of this method before considered in that in him RTT is calculated at loaded the streams of traffic networks, while measurings with the use of ping and similar methods often conducted on the unloaded network, that gives optimistic results for the values of RTT. However to take advantage of this method not succeeded: from the dispatch of packages accepted difficult character the packs of loud speaker of traffic, and linear increase of one-sided delay from a frequency change quite not traced. As a result frequency change deleted unright. It resulted in that got a method a value RTT was considerably (in two-three times) higher than that which the utility of ping reported for this network. To simplify further researches, it is recommended to apply a complex from two programs for measuring of RTT, analogical to described higher. Start programs, sending ekho-zaprosy, it is necessary to make with measuring (that when a network will be loaded) beginning, and after his completions to make the statistical analysis of the measured values RTT. In this work it was not succeeded to realize these additional measurings in connection with technical problems in work of host 'Abiturient'.

RTT usefully to the estimation of descriptions of network. If two his constituents are one-sided delays in every side - approximately equal, it is possible to utillize RTT /2 for an estimation usrednennoy one-sided delay. Comparing this value to by the critical value of network delay for this method encoding after which quality of perception of multimedia stream in interactive application falls sharply, it is possible to estimate quality of computer network.

Without the synchronized clock there is not a method to estimate, as far as near to each other of value two one-sided delays, and, means, how justly to utillize RTT /2 as an estimation one-sided delay, it is possible to take advantage of indirect by a method: building the topology map of network and estimating asymmetricness of routing. Method of construction of topology maps of network, which is here offered, it will be on the nature inexact to want of information, but example, resulted below, shows that asymmetricness it is possible with his help to discover and estimate. The map of network is built on results analysis of data, destroyed the utility of traceroute [23], at to tracing from one host to other and then back. Built a map reflects the state of network only in the moment of construction: in any other moment of time of connection between routers can to change, and a map will have to be built anew. However, routes in global networks usually or relatively stable (remain permanent at least on a few days), or constantly there is more than one route from a source to the receiver, and the passed packages are involved by all of these routes. More detailed information about descriptions of routes of global networks contained in [24]. On a fig. 3 the results of implementation are resulted traceroute on both routes. On a fig. 4 the built is presented on the basis of these information there is a map of network. [alice04]$ traceroute abiturient.stu.neva.ru traceroute to abiturient.stu.neva.ru (194.85.96.169), 30 hops max, 38 byte packets

1 wrk-ptc (195.19.225.65) 1.467 ms 1.320 ms 1.341 ms

2 Ch0.LE-PTC.spbu.ru (195.19.224.18) 3.119 ms 5.414 ms 3.280 ms

3 spb-4-gw.runnet.ru (194.190.255.157) 6.063 ms 36.462 ms 16.570 ms

4 spb-gw.runnet.ru (194.85.36.41) 131.024 ms 14.295 ms 19.970 ms

5 Rusnet-gw.runnet.ru (194.190.255.142) 12.401 ms 13.164 ms 7.990 ms

6 filter.stu.neva.ru (194.85.4.8) 12.361 ms 14.536 ms 13.515 ms

7 abiturient.stu.neva.ru (194.85.96.169) 15.062 ms 8.504 ms 7.779 ms

[abiturient]$ traceroute alice04.spbu.ru traceroute to alice04.spbu.ru (195.19.225.94), 30 hops max, 38 byte packets

1 Rusnet-spbgpu-gw.neva.ru (194.85.96.62) 0.278 ms 0.229 ms 0.162 ms

2 Int-6.100m.le-gw.rusnet.ru (194.85.4.242) 1.704 ms 1.619 ms 1.661 ms

3 Int-1.100m.le-gw2.rusnet.ru (194.85.4.12) 0.606 ms 0.632 ms 0.584 ms

4 spb-gw.runnet.ru (194.190.255.141) 2.393 ms 3.076 ms 3.678 ms

5 spb-ix.runnet.ru (194.85.36.42) 3.300 ms 3.033 ms 4.419 ms

6 ptc.spbu.ru (194.190.255.158) 11.430 ms 3.808 ms 3.956 ms

7 Ptcgate.spbu.ru (195.19.224.25) 8.756 ms 6.787 ms 7.573 ms

8 alice04.spbu.ru (195.19.225.94) 7.634 ms 21.700 ms 26.742 ms

ris. 2. These utilities of traceroute at tracing in both sides. On a fig. 4 rectangles are designated by routers, by signs question those IP-адреса of interfaces which not succeeded are marked to define from data of traceroute. The asymmetricness is observed in the segment of network of provider Rusnet. Because a difference is in a number transitions for routes 8-7=1 is equal in both sides, such the asymmetricness scarcely hinders to utillize a value RTT /2 as an estimation for an one-sided delay.

ris. 3. Topology map of network, built on the basis of information of traceroute

Traffic, generated every host, was formed as follows: a pack was generated from 10 packages for 1400 byte each, after a pause followed in 1 with. This stream is alike on traffic of Realaudio, but creates the large loading on a network. Middle bit speed makes him about 67 Kbit/sek (in each of two sides). It is enough for a transmission about 10 streams of Realaudio with by low quality of translation of speech (6,5 Kbit/sek), or about 4 streams of Realaudio with musical information in middle quality (16 Kbit/sek), or for one videostream with the sound of low qualities. Every measuring lasted 180 with. The size of package was chosen large, that it is possible it was to estimate bit speed of bottleneck networks (see further), but at possibility to conduct the more measurings it is necessary to choose the size of package, typical for an appendix which it is planned to utillize in the probed network. In addition, follows to set bit speed of stream in accordance with expected loading on a network.

For of every track of experiment was calculated followings parameters of computer network:

1. Interactive interval for the looked after values intervals between the receipts of packages from data of MGEN - calculated for both streamline and was a measure displays of 'shaking' (to the unevenness of receipt packages). Because packages were sent packs for 10 things intervals between the receipts of nearby packages in a pack were small and determined the transmission of packages on a network, and interval between the last package of pack and first package next pack made about 1 with and determined pauses between packs. Therefore all of intervals more than 0,9 with not taken into account at an analysis for simplification.

2. Stake of losses of packages during one measuring also calculated for both streamline. After calculations for every track results were represented on graphic arts as pochasovoy depen

Conclusions

     Multimedia has the most direct relation to development of internettechnologies. Became possible to send audyo- and videoreport by e-mail, and also to communicate over the Internet in real time, seeing, here, an interlocutor on the screen of computer, which quite recently was yet simple by the dream. Already a few years there are the technical decisions which settle to build the systems of transmission of multimedia-reports without the loss of quality. Even the most inexperienced user now can be without ceremony connected to the network Internet, be found, be looked over, even to listen any information interesting it from any point of world, and all of it became possible with the development multimedia

1. Anna Watson and M. Angela Sasse. "The Good, the Bad, and the Muffled: the Impact of Different Degradations on Internet Speech". University Colledge London, 2000.

2. Art Mena and John Heidemann. "An Empirical Study of Real Audio Traffic", in Proceedings of the IEEE Infocom, Tel-Aviv, Israel, Mar. 2000, USC/Information Sciences Institute, pp. 101-110, IEEE

3. H. Schulzrinne, A. Rao, R. Lanphier. Real Time Streaming Protocol (RTSP). April 1998.

4. Internet Engineering Task Force, http://www.ietf.org/

5. ITU-T P.800 Methods for subjective determination of transmission quality. http://www.itu.int/publications/itu-t/iturec.htm

6. Kun-chan Lan, John Heidemann. "Structural Modeling of RealAudio traffic". USC/Information Sciences Institute, 2001.

7. Maureen Chesire, Alec Wolman, Geoffrey M. Voelker, and Henry M. Levy. "Measurement and Analysis of a Streaming-Media Workload", 2001.

8. Network Simulator ns-2, http://www.isi.edu/nsnam/ns/

9.R.Wolff, "Poisson Arrivals See Time Averages", Operations Research, 30(2), pp. 223-231, 1982.

10.RealNetworks, "RealNetworks documentation library", http://service.real.com/help/library/

11.RFC1305. Network Time Protocol (Version 3). Specification, Implementation. D. Mills. March 1992.

12.RFC2205. Resource ReSerVation Protocol (RSVP) -- Version 1 Functional Specification. R. Braden, Ed., L. Zhang, S. Berson, S. Herzog, S. Jamin. September 1997.

13.RFC2330. Framework for IP Performance Metrics. V. Paxson, G. Almes, J.Mahdavi, M. Mathis. May 1998.

14.RFC2474. Definition of the Differentiated Services Field (DS Field) in the IPv4 and IPv6 Headers. K. Nichols, S. Blake, F. Baker, D. Black. December 1998.

15.RFC2679. A One-way Delay Metric for IPPM. G. Almes, S. Kalidindi, M.Zekauskas. September 1999.

16.RFC2680. A One-way Packet Loss Metric for IPPM. G. Almes, S. Kalidindi, M.Zekauskas. September 1999.

17.RFC2681. A Round-trip Delay Metric for IPPM. G. Almes, S. Kalidindi, M.Zekauskas. September 1999.

18.RFC777. Internet Control Message Protocol. J. Postel. Apr-01-1981.

19.RFC791. Internet Protocol. J. Postel. Sep-01-1981.

20.Vern Paxson, "Measurements and Analysis of End-to-End Internet Dynamics". University of California, Berkeley, 1997.

21.Vern Paxson, Sally Floyd. "Why we don't know how to simulate the Internet". Proceedings of the 1997 Winter Simulation Conference.

22.А.Г.Жданов, Д.А.Рассказов, Д.А.Смирнов, М.М.Шипилов. "Передача речи по сетям с коммутацией пакетов (IP-телефония)". Санкт-Петербургский государственный университет телекоммунникаций им. проф. М.А.Бонч-Бруевича. 2001 г.

23.В.Г.Олифер, Н.А.Олифер. "Новые технологии и оборудование IP- сетей". СПб.: БХВ-Петербург, 2001 г.- 512 с., ил.

24.В.Г.Олифер, Н.А.Олифер. Компьютерные сети. Принципы, технологии, протоколы. СПб: Питер, 2001. - 672 с., ил.