next up previous contents
Next: Using convertors Up: Performance prediction Previous: Performance prediction   Contents

Introduction

As we have seen in Chapter 9 the agent needs some information from the SeD to make an optimal scheduling. This information is a performance prediction of the SeD. The agent will ask the SeD to fill the data structure defined in Chapter 9 with the information it needs. The SeD returns the information and the agent can make the scheduling.
Performance prediction can be based on hardware information, the load of the SeD (the CPU load, the memory availability, etc.) or an advanced performance prediction can combine a set of basic performance predictions. Completely custom data can even be used for performance prediction, when using plugin schedulers. A performance prediction scheduler module named CoRI is available.
CoRI is described in Section 10.2.
In table 10.1 you can see which information is available by default with CoRI.


Table 10.1: Dependencies of the available information on the compiling options
always available:
TCOMP
FREECPU
FREEMEM
NBCPU
CPUSPEED
TOTALMEM
AVGFREECPU
BOGOMIPS
CACHECPU
TOTALSIZEDISK
FREESIZEDISK
DISKACCESREAD
DISKACCESWRITE
ALLINFOS
when compiled with -DDIET_USE_BATCH=ON
PARAL_NB_FREE_RESOURCES_IN_DEFAULT_QUEUE




Subsections
next up previous contents
Next: Using convertors Up: Performance prediction Previous: Performance prediction   Contents
The DIET Team - Ven 29 mai 2015 09:29:04 IST