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

Power properties incorrectly defined on platforms #45

Closed lccasagrande closed 5 years ago

lccasagrande commented 6 years ago

Hey guys, I have faced some little problems with Batsim.

First, I could not install it by following the instructions in the doc/run_batsim.md. I had some weird problems with packages that uses the ocaml_batteries and had to remove them in order to install it. It worked right out of the box after that , are these packages really needed? I'm not so sure.

After that, I tried to enable the SimGrid energy plugin using the platforms and workloads from this repo and I've faced another problem:

/tmp/nix-build-simgrid-batsim.drv-0/simgrid/src/surf/plugins/energy.cpp:210: [root/CRITICAL] Power properties incorrectly defined - could not retrieve idle, min and max power values for host master_host

I checked the platform files and the SimGrid documentation and find out that the property "watt_per_state" should have another value for the AllCores, but in the "What if the host has only one core?" section it says that it's OK to just have two values with you have only one core. I've tried to force the use of 1 core with the proper attribute, but I had an error: "Bad attribute cores in host element start tag". I've tried all the energy platforms in this repo and had the same problem.

To fix it, I had to repeat the last value in watt_per_state of each host to be able to run it using the energy plugin. It would be nice to fix it for others facing the same problem.

You did a nice work on this simulator. It will save me a lot of time, very good work.

Best regards.

mpoquet commented 6 years ago

Hello Lucas,

Thanks for contributing to Batsim! :)

Nix issue

Indeed, there was a problem in the definition of some Nix packages in datamovepkgs. ocaml_batteries was used instead of batteries, causing the failure of any installation via nix-env -i (unless -A is used, as in this case only the required descriptions are evaluated). This should be fixed in e32b955.

Energy platform definition issue

The problem you encountered probably comes from a version mismatch between SimGrid and the platform file. The platform files in Batsim's codebase (in platforms/) are usually working with their corresponding SimGrid version:

As it seems you are interested in studying energy-related phenomena with Batsim, please note that your results may be impacted by an ongoing SimGrid issue about the energy consumption of (1-core) hosts executing parallel tasks.

Best regards, Millian

lccasagrande commented 6 years ago

Yes, I'm actually developing a scheduling algorithm to minimize energy consumption and using Batsim to evaluate it.

Thanks for point me that out.

Best Regards.

mpoquet commented 5 years ago

Should we close this issue @lccasagrande?

lccasagrande commented 5 years ago

Yes for sure, thanks for the help.