You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The naive task mapping extraction code in quincy_dispatcher.cc is slowing things down at scale (most notably when running the simulator on the full Google trace) since it takes a few seconds to extract the task mappings.
We have an algorithmically superior implementation in Flowlessly, which we should back-port into the Firmament code base, so that all flow solvers benefit from it.
The text was updated successfully, but these errors were encountered:
@ICGog is this actually still an issue? I know that the dispatcher code changed significantly since, but it looks like GetMappings() in SolverDispatcher didn't change substantially. Did your recent scalability and performance improvement work would have hit the inefficient assignment extraction as an issue?
I didn't hit this issue in my scalability tests because I was using Flowlessly which can use its fast assignment algorithm to just return task mappings. However, this is likely to still be a problem if we use other solvers.
The naive task mapping extraction code in
quincy_dispatcher.cc
is slowing things down at scale (most notably when running the simulator on the full Google trace) since it takes a few seconds to extract the task mappings.We have an algorithmically superior implementation in Flowlessly, which we should back-port into the Firmament code base, so that all flow solvers benefit from it.
The text was updated successfully, but these errors were encountered: