----------------------------------------------------------------------------
-SimGrid (3.28.1) NOT RELEASED YET (v3.29 expected September 22. 2021, 19:21 UTC)
+SimGrid (3.29.1) NOT RELEASED YET (v3.30 expected December 21. 2021, 15:59 UTC)
+
+Engine:
+ - New function: run_until(date) allowing to split the simulation.
+
+SMPI:
+ - Dynamic costs for MPI operations: New API to allow users to dynamically
+ change injected costs for MPI_Recv, MPI_Send and MPI_Isend operations.
+ Alternative for smpi/or, smpi/os and smpi/ois configuration options.
+
+Documentation:
+ * SimGrid MPI calibration of a Grid5000 cluster" presenting how to properly
+ calibrate MPI communications in SimGrid.
+
+Python:
+ - Thread contexts are used by default with Python bindings. Other kinds of
+ contexts revealed unstable, specially starting with pybind11 v2.8.0.
+
+----------------------------------------------------------------------------
+
+SimGrid (3.29) October 7. 2021
+
+The "Ask a stupid question" release.
+
+We wish that every user ask one question about SimGrid to celebrate.
+On Mattermost, Stack Overflow or using the issues tracker.
+
New modeling features:
- - Non-linear resource sharing modeling resources whose performance drops with contention.
+ - Non-linear resource sharing, modeling resources whose performance heavily degrades with contention:
- The total capacity may be updated dynamically through a callback
- and depends on the idle capacity and the number of concurrent flows.
+ and depends mainly on the number of concurrent flows.
- Examples (both cpp and python): io-degradation, network-nonlinear, exec-cpu-nonlinear
- Dynamic factors: model variability in the speed of activities
- - The factor of activities can be updated dynamically through a callback
- and depends on (TODO: unsure)
- - (activities with factor=0.5 need twice the instantaneous speed for the same progression)
- - This can model variation in the progression speed over time,
- or the tasks' CPU affinity, related the "Unrelated Machines" problem in scheduling,
- or TODO: speak of debiasing?
- - This existed for network, and now for CPU and disk too.
- - For that, resources can take a callback that computes the activity factor
- (multiplicative factor applied when updating the amount of work remaining).
- - Example: examples/cpp/exec-cpu-factors
+ - Each action can now have a factor that affects its progression.
+ This multiplicative factor is applied when updating the amount of work
+ remaining, thereby an activity with factor=0.5 only uses half of the
+ instantaneous power/bandwidth it is allocated and will appear twice
+ slower than what it actually consumes.
+ - This can be used to model a overhead (e.g., there is a 20 bytes
+ header in a 480 bytes TCP packet so the factor 0.9583) but the novelty
+ is this factor can now easily be adjusted depending on activity's and
+ resources characteristics.
+ - This existed for network (e.g., the effective bandwidth depends
+ on the message in SMPI piecewise-linear network model) but it is now
+ more general (the factor may depend on the source and destination and
+ thus account to different behaviors for intra-node communications and
+ extra-node communications) and is available for CPUs (e.g., if you
+ want to model an affinity as in the "Unrelated Machines" problem in
+ scheduling) and disks (e.g., if you want to model a stochastic
+ capacity) too.
+ - For that, resources can be provided with a callback that computes
+ the activity factor when creating the action.
+ - Example: examples/cpp/exec-cpu-factors
+ - The same mechanism is also available for the latency, which
+ allows to easily introduce complex variability patterns.
Python:
- Added support to programmatic platform creation in Python.
