Job Traces

netSLS simulates Hadoop jobs given by a job trace file. Additionally information about the network topology can be provided via a topology file.

Job Trace File

Unlike the original SLS, netSLS only supports a slightly extended SLS job trace format (cf. documentation):
As described in Overview and Architecture, each Map-task requires information about its split locations. Therefore every Map-task container in the job trace has an additional parameter container.splitLocations. For example:

"container.splitLocations" : [
  "/rack1/node1",
  "/rack2/node5"
]

Topology File

Although all necessary topology information could be acquired from the job trace, SLS currently has no means to do it. Instead one can provide SLS with an additional topology file. The format is documented here.

If no topology file is provided, SLS automatically assumes a flat one-rack hierarchy. I.e. every node is in the "default-rack".

Generator

In the directory "generator", netSLS provides Python scripts for generating random artificial job traces and the corresponding topology files. The scripts do not take any input arguments; all parameters have to be changed directly in the scripts.

• generator.py generates job traces.
• generatorTopologyFile.py generates topology files.