More Network Locality (Netloc) progress
We announced the Network locality project at SC’13, and generated a LOT of interest (far more than I even anticipated!). As a refresher, here’s a link to a a blog entry we wrote about Netloc back in November.
There is still much work to be done; we’re actively continuing work in multiple areas:
- Josh and Brice just posted a bunch of documentation to the web site (this was the #1 request from all our discussions at SC’13).
- We’re investigating writing non-OpenFlow-based Ethernet readers.
- We’re investigating making the IB reader be a plugin to OpenSM (based on some nice prior work from TACC and OSU).
- We’re taking a first swipe at integrating Netloc in Open MPI.
The new readers/plugins are just icing on the cake, and are designed for end-user ease of use — the functionality was already there, it just wasn’t easy to setup and use.
The real meat of the work is integrating into a research fork of Open MPI (yay git for easy forking and keeping up with the head of Open MPI development!). Using the Netloc data is a major reason why this project useful, after all.
As we integrate the Netloc library into Open MPI, we’re finding the strengths and weaknesses in our initial design. We’re feeding the knowledge gained back into Netloc itself, improving it as we go along. This is a major benefit of software co-design between multiple projects.
We’re still very much interested in your feedback; what do you want to see out of Netloc? What kind of queries do you want to run? How would you anticipating using Netloc data? And so on.
Join our users mailing list and let us know!