V dobi klasičnih telefonskih omrežij je bilo porajanje prometa porazdeljeno po Poissonovi porazdelitvi. V sodobnih omrežjih, kot je internet z zanj tipičnim raznolikim prometom, pa to ne drži več. Vse bolj se zdi, da je dolgorepa ali Paretova porazdelitev tista, po kateri se porazdeljujejo medprihodni časi paketov.
V okviru delovne hipoteze se sprašujemo, kako se porazdeljujejo medprihodni časi paketov video prometa, ki se pretaka po omrežju. Določiti želimo, ali je porazdelitev medprihodnih časov bolj podobna eksponenti porazdelitvi, ali pa Paretovi porazdelitvi.
V diplomski nalogi analiziramo eksperimentalno pridobljeni spletni promet med izvorom in ponorom in poskušamo ugotoviti kateri porazdelitvi je porazdelitev medprihodnih časov bolj podobna. V okviru diplomske naloge smo postavili dva tipa eksperimentov, pri čemer je v obeh prisoten video strežnik, ki oddaja video promet (oddajna naprava) in računalnik, ki promet sprejema (sprejemna naprava). Eksperimenta se razlikujeta v fizični lokaciji oddajne naprave. V prvem tipu eksperimenta se oddajna naprava nahaja v domačem omrežju, v drugem tipu eksperimenta pa v Združenih državah Amerike. Analiza je potekala tako, da smo generirali takšno umetno Paretovo porazdelitev in takšno umetno eksponentno porazdelitev, ki sta najbolj podobni eksperimentalno pridobljeni porazdelitvi. Nato smo vsako umetno generirano porazdelitev primerjali z eksperimentalno pridobljeno porazdelitvijo s pomočjo Jensen-Shannonove divergence. Koda vseh skript, ki smo jih kreirali za pomoč pri analizi, je dostopna na naslovu
https://github.com/JeznaSpianca/diploma_skripte. Rezultati, ki smo jih pridobili, potrjujejo našo hipotezo o porazdelitvi omrežnega prometa.
| Language: | English |
|---|
| Title: | Analysis of packet interarrival times distribution in video streaming traffic |
|---|
| Abstract: |
|---|
In the age of classical telephone networks, the generation of traffic was distributed according to the Poisson distribution. In modern networks, such as the Internet with its typical diverse traffic, this is not the case. Increasingly, the long-tailed or Pareto distribution seems to be the one by which the inter-arrival times of packets are distributed.
In the context of working hypotheses, we wonder how the inter-arrival times of video traffic packets flowing across the network are distributed. We want to determine whether the inter-arrival time distribution is more similar to the exponential distribution or to the Pareto distribution.
In this thesis we analyze the experimentally obtained web traffic with source and sink and try to determine which distribution is more similar to the distribution of inter-arrival times. As part of this thesis, we set up two types of experiments, both of which have a video server that transmits video traffic (transmitting device) and a computer that receives (receiving device). The experiment differs in the physical location of the transmitting devices. In the first type of experiment, the transmitting device is located in the home network, and in the second type of experiment, in the United States. The analysis was performed by generating such an artificial Pareto distribution and such an artificial exponential distribution that is most suitable for the experimentally obtained distribution. We then compared each artificially generated distribution with the experimentally obtained distribution using the Jensen-Shannon divergence. The code of all the scripts we created to help with the analysis is available at https://github.com/JeznaSpianca/diploma_skripte. The results we obtained confirm our hypothesis about the distribution of network traffic.
|
| Keywords: | traffic distribution, Poisson distribution, Pareto distribution, experiment, exponential distribution, Jensen-Shannon divergence |
|---|
