It just suddenly occurred to me that, with remote machines, an obvious throughput increase could be achieved by splitting cycles into local-bound and remote-bound parts, and having the local-bound part of cycle n+1 (perturbing, fuzzing, lifting, possibly copying) spin off while waiting for the remote-bound part of cycle n (compiling, running, copyback, analysis) to finish. This would fill up gaps in remote machine usage caused by it waiting for the next
One difficulty would be working out how to show the two activities happening simultaneously on the dashboard. Maybe separate local and remote sections.
It just suddenly occurred to me that, with remote machines, an obvious throughput increase could be achieved by splitting cycles into local-bound and remote-bound parts, and having the local-bound part of cycle
n+1
(perturbing, fuzzing, lifting, possibly copying) spin off while waiting for the remote-bound part of cyclen
(compiling, running, copyback, analysis) to finish. This would fill up gaps in remote machine usage caused by it waiting for the nextOne difficulty would be working out how to show the two activities happening simultaneously on the dashboard. Maybe separate local and remote sections.