[codes-ross-users] updates on workload support

Jenkins, Jonathan P. jenkins at mcs.anl.gov
Wed Dec 10 15:29:32 CST 2014


Hi all,

I’ve had a bit of time over the past few days to make some progress on the IO/network workloads APIs of CODES. In particular, I have experimental branches on both codes-base and codes-net repositories (both named workload-combine) that integrate the network workload API and parsers into the IO workload API. In other words, the directory “codes-base/src/network-workload” is now a dead directory, with relevant content having been moved to “codes-base/src/workload”.

At some point this will be pushed to master, but there are some folks who are using the current setup on master that may need to put in fixes or other changes, so I’m holding off on merging for the time being.

Some notes:
- the scalatrace workload generator is only a partial implementation, and won’t work for simulations (it assumes one reader per MPI rank, whereas multiple LPs can be calling into workload functions on the same MPI rank). The dumpi generator, another MPI tracer (http://sst.sandia.gov/about_dumpi.html), is more complete for this purpose.
- the existing workload generators need some cleaning/refactoring. They currently are not too pretty.
- I also need to update the documentation, but the usage is near identical to the original IO workload API.
- there aren’t any generators yet that generate both IO and network events (contributions here would of course be welcome :) ).

Thanks,
John


More information about the codes-ross-users mailing list