Home » SSTDocumentation » SST V14.1.0 Release Notes

SST V14.1.0 Release Notes

SST V14.1.0 Release Notes
Fixed issues and significant code changes:

General

This release has several major improvements to checkpoint/restart support. See the guide here.
- Statistics can be checkpointed
- Checkpoint/restart is supported for parallel execution
- Checkpoint can be triggered on real (wall) time intervals in addition to simulation time
14.1 also introduces new experimental SST-Core features. These are prototypes that will be improved in future releases. See SST-Core release notes for details.
- Interactive debugging. The simulation can be dropped into an interactive mode where the state of components/subcomponents/etc. can be observed and changed. Only serializable (checkpointable) components can be introspected in this way. See sst --help for more information.
- User-defined signal actions. SST’s default signal handling has been extended to allow element libraries to define their own signal actions for SIGUSR1, SIGUSR2, and SIGALRM. A number of optional handlers are also provided with SST-Core. See sst --help for more information.
The ariel, mercury, maskMPI, and simpleElementExample element libraries have new capabilities; bug fixes and minor improvements were made to other libraries.

Deprecation and Removal Notices

The test framework (sst-test-core and sst-test-elements) has some newly deprecated functions. The framework is moving towards using Python functions in favor of custom functions where Python has equivalent capability. The following are deprecated.
- testing_is_PIN2_used(). Pin2 is no longer supported in SST.
- host_os_get_system_node_name(). Use platform.node() instead.
- host_os_get_kernel_type(). Use platform.system() instead.
- host_os_get_kernel_release(). Use platform.release() instead.
- host_os_get_kernel_arch(). Use platform.machine() instead.
- host_os_is_osx(), host_os_is_linux(), host_os_is_centos(), host_os_is_rhel(), host_os_is_toss(), host_os_is_ubuntu(), host_os_is_rocky(). Use host_os_get_distribution_type() or host_os_get_distribution_version() instead.
- host_os_get_num_cores_on_system(). Use multiprocessing.cpu_count() instead.
- sst_core_config_include_file_get_value_int, sst_core_config_include_file_get_value_str(). Use sst_core_config_include_file_get_value() instead.
- sst_elements_config_include_file_get_value_int, sst_elements_config_include_file_get_value_str(). Use sst_elements_config_include_file_get_value() instead.
- sstsimulator_conf_get_value_str(), sstsimulator_conf_get_value_int(), sstsimulator_conf_get_value_float(), sstsimulator_conf_get_value_bool(). Use sstsimulator_conf_get_value() instead.
- initializeClass() is deprecated in favor of UnitTest’s setUpClass().

These notices were provided in SST 14.0:

Removed previously deprecated functions: Link::sendInitData(), Link::recvInitData(), SimpleNetwork::sendInitData(), SimpleNetwork::recvInitData() and BaseComponent::getSimulation() have been removed. The simulation.h header file is also removed.
- See the new deprecation history page for suggested alternatives
The SST_ELI_REGISTER_SUBCOMPONENT_DERIVED and SST_ELI_REGISTER_MODULE_DERIVED macros are removed
- See the new deprecation history page for suggested alternatives
The SimpleMem interface is removed. Use the StandardMem interface instead.
The Event::Handler and Clock::Handler handler types are deprecated in favor of Event::Handler2 and Clock::Handler2. The new types support checkpoint/restart. See checkpoint guide for details. Eventually the Handler names will be reintroduced to match the new Handler2 types.

Known Issues

The OpenMPI 4.1.1 available through Rocky Linux and other OS distributions masks SST’s error return codes which causes some SST tests to fail. A workaround is to duse OpenMPI 4.1.6.

SST-Core

Added capability to generate checkpoints of parallel simulation. This changes the output format of checkpoints for all simulations to a multi-file format as opposed to the initial single file checkpoint in SST 14.0.
Added checkpoint support to Statistic and StatisticOutput types
Added an experimental interactive mode where the simulation can be dropped into an interactive console. Interactive mode can be trigger by setting a time on the command line, or through API calls in BaseComponent and RealTimeAction. The console used for the interactions can be specified on the command line and are user creatable by inheriting from the InteractiveConsole class. Interactive mode has only been tested in serial runs and may not yet function as anticipated in parallel. The interactive API will continue to involve.
Added a debugging capability prototype using interactive mode. This works in serial simulations only and has a rudimentary console debugger. In interactive mode, Component state can be queried and changed and the simulation can be stepped forward in simulation time units. Expect improvements in future releases.
Added default verification policy to SharedSet verification policy
Fixed issues with mutex_read and mutex_find in all SharedRegion classes
Added RealTimeAction class for element-library defined signal actions and made signal actions configurable on the command line.
Added real-time heartbeat option in addition to existing simulation-time heartbeat.
Added real-time checkpoint option in addition to existing simulation-time checkpoint.
Test framework improvements and clean up
- Type annotations added
- Duplicate functions deprecated in favor of Python or more generic ones
- Added ability to provide a timeout other than default in os_cmd() calls
- Fixed bug in sstsimulator_conf_get_value_bool and sstsimulator_conf_get_value when parsing boolean values
- Updated parsing functions to accomodate OpenMPI 5.0.3 output format.
- Fixed bug in ncurses detection
- Made function testing_check_is_scenario_filtering_enabled internal as it should not be used by test suites
Fixed bug when running with --run-mode=init
Fixed issue where params where serialized but the keyset was not saved for checkpoint
Modified TraceFunction to respect the environment variable SST_TRACEFUNCTION_ACTIVATE for controlling whether output is enabled and sst_TRACEFUNCTION_INDENT_MARKER to control output indent format.
Added quit command to sst-info
Fixed issue with correctly parsing options passed to profile points on the command line

SST-Elements

Ariel
- Added support for running MPI applications on Ariel CPUs
- Added ariel_disable in the Ariel API to allow for tracing to be turned off
- Updated the internal Ariel API to include all functionality from the API found in sst-tools
Ember
- Fixed support for OTF2 including bcast, scatter, allgather, and alltoall collective operations. Also added addCompute parameter to add ember compute time between MPI calls.
Kingsley
- Fix to correctly delete credit events
MaskMPI
- Full collective support
MemHierarchy
- Fix to correctly delete dropped prefetch events
- Fix to correctly delete init events in the MemNIC
- Fixed coherence protocol bug in noninclusive shared cache
- Added ability to skip a potentially expensive memory address overlap check in MemNIC during init
Mercury
- Proper rank mapping via Merlin
- Fix to event serialization
Merlin
- Added ability to put a user subcomponent on the merlin bridge’s linkcontrol
- Fix to correctly delete init events
SimpleElementExample
- Made example elements checkpointable and added checkpoint example
Vanadis
- Minor test updates

SST-Macro

None

Fixed issues and significant code changes:

SST-Core

1089 Base comp ser

1092 Update checkpoint branch - statistics

1094 Change to lookup stat field ID on restart

1097 Make HDF5 output checkpoint compatible

1101 Fix test for when hdf5 is not present

1102 Move output serialization

1103 Header file cleanup in output.h and output.cc

1093 Modified TraceFunction object to require that an environment variable be set to enable output. This makes it possible to leave the TraceFunction calls in while developing, but still be able to run and pass testing. Update TraceFunction for better formatting. (#1093)

1104 Add SST-Info interactive test, quit command

1106 Fix sst info tests

1107 Reorganize serialization code in preparation for larger changes

1110 ncurses detection fixes

1108 Unify signal handling infrastructure

1111 deprecate initializeClass() method

1113 Update GitHub Actions versions

1112 Remove unused tests/test_RNGComponent.py

1115 Realtime bug fix and other fixes

1119 factor out reading and filtering lines from test output

1114 Remove redundant initializeTestModule_SingleInstance() methods

1121 fixed sstsimulator_conf_get_value_bool

1127 Update and pin version of clang-format-lint-action

1128 Python 3.11: Update datetime command for current UTC time

1130 Fix RealTime tests that were sending signals too quickly for the slower test platforms

1125 TarFile.extractall() now warns when not specifying extraction filter

1133 Loosen test criteria for realtime tests to accomodate time uncertainty

1132 Initial implementation of parallel checkpointing.

1136 deprecate testing_is_PIN2_used()

1137 deprecate simple host_os_get_* methods

1144 Add handling for output file naming conventions in OpenMPI 5.0.x

1145 add deprecation warnings for various functions

1143 Make signal handlers user-settable.

1148 make helper function internal

1152 fix call to os.path.isfile()

1150 override default timeout in os_cmd() calls

1154 remove calls to deprecated functions

1155 Give SharedSet a default verification policy

1158 Bug fixes for parallel checkpointing.

1151 Add initial set of Python type annotations

1159 Serialization mapping mode and interactive mode

1160 Fix test logic for shared array

1162 Fixes uninitialized object segfault in debugger

1161 Fix building with –enable-perf-tracking

SST-Elements

2407 merlin/bridge: load user subcomponent for networkIF (merlin)

2406 Add basic test for simpleElementExample checkpointing (simpleElementExample)

2401 Update simpleElementExample for checkpointing (simpleElementExample)

2398 Resolve some memorypool undeleted events (mutiple)

2394 mask-mpi: add missing collectives (maskMPI)

2392 Ember OTF2 linking and Bcast fix (with an added test) (ember)

2389 Mercury: rankmap (mercury)

2388 Fix coherence protocol bug in noninclusive shared cache (memHierarchy)

2386 Ariel MPI (ariel)

2384 Remove redundant initializeTestModule_SingleInstance() methods (multiple)

2381 Fixes for Vanadis Config Scripts (vanadis)

2379 remove initializeClass() method (multiple)

2377 Mercury: fix serialization headers (mercury)

2376 mem_nic range check enable/disable and test (memHierarchy)

SST-Macro

717 remove initializeClass() method