In recent years there has been a sharp increase in access to internet, causing data centers (DC) should dynamically adapt its network infrastructure to face possible problems of congestion, which is not always given in a timely manner. Given this, new network topologies have been proposed in recent years as a way to provide better conditions for handling internal traffic, however it is common for the study of these improvements the need to recreate the behavior of a real DC in models of simulation/emulation. Therefore it becomes essential to validate these models, in order to obtain consistent results with reality. Such validation is possible through the identification of certain properties which are derived from the variables and parameters that describe the network and are maintained in the DC topologies for different scenarios and/or configurations. These properties, known as invariant, are an expression of the operation of the network in real environments, such as the longest path between two nodes or the minimum number of links that must fail before a loss of connectivity on one of the nodes of the network. In this paper, the identification, formulation and testing of two invariants for the Fat-Tree topology is performed, using MiniNet as the software emulator. The conclusions show good agreement between the analytical and the practical.
|Translated title of the contribution||Determination of Invariants in Large Data Centers Based on Fat-tree Topology|
|Original language||Spanish (Ecuador)|
|Number of pages||6|
|State||Published - 27 Feb 2015|
- Network invariants
CACES Knowledge Areas
- 216A Network and Database Design and Administration