Ing. Ondřej Votava

Síťový tým na katedře telekomunikační techniky FEL ČVUT v Praze, se snaží přispět vývojem a rozšiřováním softwarového nástroje FlowPing. Vychází přitom z ověřených principů pro měření a přináší komplexnější možnosti diagnostiky datových spojů. Otevřený nástroj FlowPing umožňuje vyvíjet a používat i nové metodiky pro testování parametrů komunikační sítě na vrstvě L4. Usnadňuje tak validaci přenosového prostředí dle konkrétních nároků koncových služeb.

Metodika testování přístupových sítí je založena na dokumentech BEREC a ČTÚ. Lze předpokládat, že se dále budou zpřísňovat doporučené parametry v souvislosti s výhledem na tzv. sítě s velmi vysokou kapacitou v horizontu roku 2025. Přístup k Internetu tedy už dnes není negarantovanou „best effort“ službou. Proto také vzrůstá potřeba ověřovat parametry přípojek. V tomto článku jsou představeny základní metodické přístupy a nástroje, které na základě našich zkušeností doporučujeme k praktickému využití. Nad otevřenými nástroji, jako je iPerf a FlowPing jsme vyvinuli nadstavbovou platformu, která usnadní vyhodnocení testů, zvláště těch, které používají více různých toků jak souběžně, tak i v postupných po sobě následujících krocích.

Heterogeneous computational platform offers a great ratio between the computational power and the price of the system. Static and dynamic scheduling methods offer a good way of how to use these systems efficiently and therefore many algorithms were proposed in the literature in past years. The aim of this article is to present the dynamic (on-line) algorithm which schedules multiple DAG applications without any central node and the schedule is created only with the knowledge of node’s network neigbourhood. The algorithm achieves great level of fairness for more DAGs and total computation time is close to the standard and well known competitors.

Distributed computing may be looked at from many points of view. Task scheduling is the viewpoint, where a distributed application can be described as a Directed Acyclic Graph and every node of the graph is executed independently. There are, however, data dependencies and the nodes have to be executed in a specified order. Hence the parallelism of the execution is limited. The scheduling problem is difficult and therefore heuristics are used. However, many inaccuracies are caused by the model used for the system, in which the heuristics are being tested. In this paper we present a tool for simulating the execution of the distributed application on a "real" computer network, and try to tell how the execution is influenced compared to the model.

Smart mobile phones, tablets, notebooks, all these devices require mobile connection to the Internet. Mobile Internet providers promise the great connection parameters and great speeds. The reality often differs and several applications, which require permanent connection to the Internet may not work correctly, which leads to the users' disappointment. The wrong behavior could be fixed in most cases if the programmer tested the application in the specific conditions. We present a tool for analyzing mobile data networks, focused on data transfer, and the emulation tool which is able to emulate the mobile connection on a local network which is a common development environment.

Distributed computing may be looked at from many points of view. Task scheduling is the view, where the distributed application can be described as a Directed Acyclic Graph and every node of the graph is executed independently. There are, however, data dependencies and the nodes have to be executed in a specified order and hence the parallelism of the execution is limited. The scheduling problem is difficult and therefore heuristics are being used. Yet, there are many inaccuracies caused by the model used for the system, where the heuristics are being tested. In this article we present the tool for simulation of the execution of the distributed application on a "real" computer network and try to tell how the execution is influenced compared to the model.

Task scheduling is a way of how to run ``any'' application in a distributed computation system. Traditional attitude is loosely coupled with the well known structure of the application and all the data this application needs to work with is already known when the computation begins. Similarly, these data are available to all of the computation nodes. In this article we modified this model by adding constraints that some data are available only after a specific time and at a specific place. We present a modified HEFT, CPOP and DLMDAG algorithms so that they are able to take this modification into account.

Heterogeneous computational platform offers a great ratio between the computational power and the price of the system. Research in static and dynamic scheduling has been done and many algorithms were proposed in the literature. The aim of this article is to present the dynamic (on-line) algorithm which schedules multiple DAG applications dynamically without any central node and the schedule is created only with the local knowledge of the network. This algorithm does not beat the minimal makespan but offers good ratio of fairness for multiple DAGs and usage of computational resources.