Conversation
Hotfix cmake ross versioning
…fic timestamp What is left is copying the mechanism from optimistic onto conservative and optimistic realtime
This is a fix for a problem that only exists because the arbitrary function can schedule (or change) events to happen after the simulation is set to end.
Also, making sure that sequential and optimistic track the same kind of stuff (both track event processing and commit time).
Logs from some LPs get truncated with the current setup. This commit fixes the issue by creating separating log files for LPs of each PE for easier debugging. Change name of ross_lp_logs directory Changed formatting of tw_all_lp_stats Modified mkdir logic for tw_all_lp_stats abc
Create separate logs per PE for tw_lp_all_stats
This slip up makes it so that ROSS won't compile if -Wimplicit-function-declaration is treated as an error! In my system, an implicit declaration is treated as a warning and everything just compiled. It didn't in other systems.
It was awfully outdated (10 years). Hopefully the next update won't be so far in the future.
Changed every ocurrence of "arbitrary function" for "gvt hook" (it is hopefully more readable now). Simplified the interface so that there is only one interface for the gvt hook, when the tiebreaker mechanism is active or not.
This commit adds an option of passing arguments to ROSS via the command line to a .txt config file containing the same command line arguments.
This commit adds instructions to use the args-file command line argument.
If an event is scheduled passed the `end` of the simulation, the event is immediately discarded (aborted), yet it will alter the random generator for the tiebreaking mechanism (core_rng). This commit fixes that corner case and guarantees that the behavior of two simulations is always deterministic :fingerscrossed:.
With this new mode, one can test if all event reverse handlers have been properly implemented. It will run each event twice, rollbacking it once, and it will check that: (1) processing and rollbacking an event will produce the same LP state as not processing the event at all, and (2) processing the event after rollbacking it is the same as just processing the event once Additionally to checking the LP state, we also check the state of random generators. The runtime penalty of processing all events twice is an increase of at least 100%. In phold experiments, this mode takes about 2.8x times the original.
Instead of copying the full struct of the tie-breaker every single time we come across it, we can simply copy the values we care. We can shave a significant amount of time if the work performed by each LP is small compared to network time.
… process In the past, we would finish the simulation if there are no more events to process. This is not true with a GVT hook. A GVT hook call can add events to the queue even when there aren't any left.
This merge brings in four major changes: - We add a new mode --args-file which allows us to pass some arguments in a file. So, > mpirun -np 3 my-bin --synch=3 --args-file=path-to-file.txt is now possible instead of > mpirun -np 3 my-bin --synch=3 --argument-1=val --argument-2=val \ > --argument-3=val --argument-4=val --argument-5=val - We present a stable function hook at GVT. With this, we can now execute arbitrary code at GVT, which can be used for changing global parameters of the model at specific points in time, or save the model somewhere, or do whatever needs to be done globally in a consistent state - We added a new handy functions that can be used inside the GVT hook, the tw_scheduler_rollback_and_cancel_events_pe. If we call this function within the hook, we can force every PE to rollback and to remove cancel events. This helps us to keep a clean incoming queue/events to be processed queue, so that we can inspect the events that will be executed next - We added a new synchronization strategy (strategy 6, reverse handler check). It allows us to check for errors on the reverse handler (aka, the reverse handler not reversing the LP state as it should)
All tests run properly now
Member
Author
|
Bypassing rules and accepting changes. Sadly, Travis CI has not been enabled to work with CMake |
Member
|
FYI I'll be working on fixing CI when we (Kitware) are able to start working on this again (soonish). I had noticed during the Phase I of our project that CI no longer seemed to be working. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bug fixes:
With these bug fixes, ROSS can guarantee deterministic execution of a parallel optimistic simulation (the tiebreaker mechanism has to be enabled; it is enabled by default).
New features:
tw_run()to execute a GVT hook function 101f18f