search

oar-team / batsim

Batsim: Infrastructure simulator for job and I/O scheduling
GNU Lesser General Public License v3.0
30 stars 15 forks source link

Consumed energy output: instantaneous power might be cleaner #29

Open mpoquet opened 7 years ago

mpoquet commented 7 years ago

In exported file _consumed_energy.csv, the written instantaneous power on a line corresponds to the energy difference with the previous line divided by the time difference with the previous line.

This leads to power values of NA for the first power value, or when two lines are at the exact same time.

Enhancement fix: the current consumed power could be retrieved from SimGrid to match the simulation current consumption. This way a line meaning would be present focused and future oriented, instead of past-focused and past-oriented. This way, no NA value would exist.

mpoquet commented 7 years ago

Commit 11df8f4 adds wattmin information from SimGrid as a temporary fix.

Using the current power computed by SimGrid should be better, but it does not seem possible with current forked SimGrid version. @Shurakai's new energy plugin seems great to get it easily. Is it usable from MSG applications ?

mickours commented 6 years ago

Does S4U solve this issue?

mpoquet commented 6 years ago

Retrieving such information from SimGrid is not as easy as it seems.

Batsim is most of the time called at the end of a job, so the involved hosts are idle in the current power consumption. We could add a callback at the end of each computation (via s4u) but I feel that this is not really needed in our case.

Furthermore, the semantic of such power values in the _consumed_energy.csv is not so clear.
Only keeping time/energy measures might be easier to understand for users and sounder to use for analysis and visualization tools. For example, evalys only reads time/energy measures and retrieves the power information from it (figure below).

evalys power

I feel like this issue should be closed.