Restart Research E[X|t] for various distributions:

Truncated distributions in RESTART (removes tail)

Restart w/ checkpoint before and after node (as opposed to after)

Mean system time as a function of Beta {mean, variance, asympotic}

Outline

Background - define problem and terms (definitions)

Restart with a job of length of t Restart with a job of length of t split into two Restart with a job of length of t into K equal pieces [reduces tail] Restart with a job of length of t into K random (exp) sized pieces [reduces tail] Restart with a job of length of t in fixed time interval [breaks tail]

=

Structured modeling

Restart with a job of length of t in Markov Model

Checkpointing

${\tilde {a}}$

Checkpointing w/ overhead

Investigate fail rates per node as opposed to per system (B as a diagonal matrix) Investigate truncated distributions for modeling and tail clipping