memory characterizations

  • complete
  • dumps statistics on blocks that have changed (based on checksum across all blocks)
  • Applications
    • CTH, Sage, LAMPS, POP, others?
    • Issues with proper datasets
      • datasets too small (current demo apps run in 15 minutes)
      • real datasets are classified
    • We can't just give library to user (needs a modified PCT to work)
    • Courtenay may be able to run larger-scale tests
    • May not need large-scale tests (emulate strong scaling).
  • Rsqual is the primary testbed.

Other stuff:

  • New definition for optimal checkpoint interval (Laros)
    • Checkpoiont when number of changes exceed a threshold (also need to use some time-related info)
  • Will checksum give us an indication of how much work is done?
  • Do we want to have a smaller granularity of changes (smaller than block)?
    • This might be hard given that we're using hashes to identify modified blocks
    • what is the right hash function (md5, CRC, ...)?
      • Not really worried about false positives. We can't have false negatives.
      • Use lightweight hash to determine when to checkpoint, use heavyweight (accurate hash) when ready to checkpoint.

CIFTS Update

  • They want us to come up with use-case scenarios (what do we want from FTB to make us use it).
    • What features could they provide that would help with our resilience efforts?
  • They are not actually correlating information.

kitten process migration

  • No progress yet. Pedretti is on the hook.

network quiescence

  • Still working out issues with test cages.


  • Laros figured out how to shutdown a link.