Sie haben Javascript deaktiviert!
Sie haben versucht eine Funktion zu nutzen, die nur mit Javascript möglich ist. Um sämtliche Funktionalitäten unserer Internetseite zu nutzen, aktivieren Sie bitte Javascript in Ihrem Browser.

CRC 901 – On-The-Fly Computing (OTF Computing) Show image information

CRC 901 – On-The-Fly Computing (OTF Computing)


Fifth SFB 901 Seminar in winter semester 2020/2021

On December 9, 2020, the 5th SFB 901 seminar in the winter semester will take place.

4:00 - 5:00 subproject C4
Speakers: Dr. Marten Maack + Simon Pukrop
Title: Scheduling Composed Jobs in Compute Centers Enhanced by Cloud Resources

The talk will be held via BigBlueButton.


The subproject C4 deals with resource allocation and scheduling for OTF services, that is, the question
how available (computational) resources may be used to execute given OTF services in a most efficient
way. These services are usually composed of several interlinked, interacting components. Ideally, some
information on resource consumption of each composed part and on the requirements and implications
of the interactions between the parts are know. Examples include processing times, memory, bandwidth
or communication delays. One of the main focuses of this project is to investigate different types of
composite services and to use the additional information as well as possible.

When we look at the execution of composed services there are plenty of additional restrictions one might
have to take into account. In particular, we consider a scenario in which services are presented as directed
task diagrams and must be completed by a given deadline. However, parts of the composed services might
exceed the capabilities of our local machine. This might happen, for instance, in machine learning applications,
especially in the training step. On the other hand, even if each part could be performed on the local machine
in principle, the overall computational load could be simply too big to handle. In today's time, a plausible
approach to this problem is to rent additional cloud computing time in order to unburden the local system.
Hence, in our model an arbitrary number of cloud machines can be used, but we have to cover costs that
are proportional to the total computing time in the cloud. The objective of the scheduling problem is to find
a feasible assignment of each part of the composed services minimizing the overall cost. We formally define
the model an show first theoretical results.

The University for the Information Society