Directions for Future Work

Jumpshot is still evolving. We are working on enhancements in a number of areas.

• We would like to bring Jumpshot closer to completely satisfying the requirements in the list in Sections Requirements for a New System and How It Works . It meets most of these goals, as can be seen by the features discussed in this paper, but others are either missing or incomplete. In the near term we will be adding support for filtering by MPI communicator. The desirable feature of tracking events back to source code is probably much further away, because of the difficulty of achieving this in a portable way. The problems are more with capturing the information than with displaying it.
• Although Jumpshot and CLOG represent advances over upshot, nupshot, and ALOG in terms of handling large log files and complex displays, there is still much more to do. A step in this direction is a collaboration with IBM Research to develop a scalable logfile format (SLOG), capable of dealing with (and helping Jumpshot deal with) multi-gigabyte logfiles.
• The identification of states that last ``too long'' relies for the present on simple statistical measures such as the standard deviation. We would like to employ more sophisticated statistical techniques to identify ``anomalous'' parts of the run, as well as more sophisticated models.
• The current collection of displays have served us well, but as data volume increase, more elaborate visualization techniques will be needed. This is a research topic for the (near) future.