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
which lists the possible parameter names, allowed values, and brief descriptions.
3606
3606
More detailed descriptions of particular parameter values are found in the documentation in previous chapters.
@@ -3655,6 +3655,8 @@ The allowed parameter types are:
3655
3655
| group\_connections (int) | No default | Positive int | For cascase ir dragonfly, the number of intergroup connections on each switch in a Dragonfly group |
3656
3656
| redundant (vector of int) | vector of 1's | Positive ints | For Cartesian topologies (hypercube, cascadem, dragonfly, torus) this specifies a bandwidth (redundancy) multiplier for network links in each dimension. |
@@ -3778,6 +3780,7 @@ All other parameters can be filled in from `node.nic.injection`.
3778
3780
| credits (byte length) | No default || The number of initial credits for the component. Corresponds to an input buffer on another component. In many cases, SST/macro can compute this from other parameters and fill in the value. In some cases, it will be required. |
This is a series of namespaces `app1`, `app2`, and so on for each of the launched applications. These should be contained within the `node` namespace.
3798
3802
3799
-
3800
-
| Name (type) | Default | Allowed | Description |
3801
-
|-------------|---------|---------|-------------|
3802
-
| name (string) | No default | parsedumpi, cxx\_full\_main, cxx\_empty\_main | The name of the application to launch. Very few applications are built-in. Registration of external apps is shown starting in Section [3.5](#sec_tutorial_basicmpi). |
3803
-
| size (int) | No default | Positive int | The number of procs (MPI ranks) to launch. If launch\_cmd given, this parameter is not required. |
3804
-
| start (int) | 0 || The time at which a launch request for the application will be made |
3805
-
| concentration (int) | 1 | Positive int | The number of procs (MPI ranks) per compute node |
3806
-
| core\_affinities (vector of int) | Empty |||
3807
-
| launch\_cmd (string) | No default | Valid aprun or srun | This uses a launch command as would be found with ALPS or SLURM launchers on real systems, e.g. aprun -n 4 -N 1 |
3808
-
| indexing (string) | block | block, random, cart, node\_id, coordinate | The indexing scheme for assign proc ID (MPI rank number) to compute nodes |
3809
-
| node\_id\_mapper\_file (filepath) | No default || If using Node ID indexing, the file containing the node ID index list |
3810
-
| random\_indexer\_seed (long) | System time || The seed to use for a random allocation. If not specified, system time is used. |
3811
-
| allocation (string) | first\_available | first\_available, random, cart, node\_id, coordinate | The scheme to use for allocating compute nodes to a given job. |
3812
-
| random\_allocation\_seed (long) | System time || For random allocation policy. If unspecified, system time is used as the seed. |
3813
-
| node\_id\_allocation\_file (filepath) | No default || If using Node ID allocation, the file containing the list of node IDs to allocate for the job |
3814
-
| dumpi\_metaname (filepath) | No default || If running DUMPI trace, the location of the metafile for configuring trace replay |
3815
-
| coordinate\_file (filepath) | No default || If running using coordinate allocation or indexing, the path to the file containing the node coordinates of each proc (MPI rank) |
3816
-
| cart\_sizes (vector of int) | No default || Launch a contiguous block of nodes in a Cartesian topology. This gives the size of each dimension in the block. |
3817
-
| cart\_offsets (vector of int) | No default || Launch a contiguous block nodes in a Cartesian topology. This gives the offset in each dimension where the block begins. |
3818
-
| parsedumpi\_timescale (double) | 1.0 | Positive float | If running DUMPI traces, scale compute times by the given value. Values less than 1.0 speed up computation. Values greater than 1.0 slow down computation. |
3819
-
| parsedumpi\_terminate\_percent (int) | 100 | 1-100 | Percent of trace. Can be used to terminate large traces early |
3820
-
| host\_compute\_timer (bool) | False || Use the compute time on the host to estimate compute delays |
3821
-
| otf2\_metafile (string) | No default | string | The root file of an OTF2 trace. |
3822
-
| otf2\_timescale (double) | 1.0 | Positive float | If running OTF2 traces, scale compute times by the given value. Values less than 1.0 speed up computation. Values greater than 1.0 slow down computation. |
3823
-
| otf2\_print\_mpi\_calls (bool) | false || Print MPI calls found in the OTF2 trace |
3824
-
| otf2\_print\_trace\_events (bool) | false || Debugging flag that printsindividual trace events (which includes details such as when an MPI call begins, ends, and when a collective begins and ends |
3825
-
| otf2\_print\_time\_deltas (bool) | false || Debugging flag that prints compute delays injected by the simulator |
3826
-
| otf2\_warn\_unknown\_callback (bool) | false || Debugging flag the prints unknown callbacks |
3803
+
\openLongTable
3804
+
\hline
3805
+
name (string) & No default & parsedumpi, cxx\_full\_main, cxx\_empty\_main & The name of the application to launch. Very few applications are built-in. Registration of external apps is shown starting in Section [3.5](#sec_tutorial_basicmpi). \\
3806
+
\hline
3807
+
size (int) & No default & Positive int & The number of procs (MPI ranks) to launch. If launch\_cmd given, this parameter is not required. \\
3808
+
\hline
3809
+
start (int) & 0 & & The time at which a launch request for the application will be made \\
3810
+
\hline
3811
+
concentration (int) & 1 & Positive int & The number of procs (MPI ranks) per compute node \\
3812
+
\hline
3813
+
core\_affinities (vector of int) & Empty & & \\
3814
+
\hline
3815
+
launch\_cmd (string) & No default & Valid aprun or srun & This uses a launch command as would be found with ALPS or SLURM launchers on real systems, e.g. aprun -n 4 -N 1 \\
3816
+
\hline
3817
+
indexing (string) & block & block, random, cart, node\_id, coordinate & The indexing scheme for assign proc ID (MPI rank number) to compute nodes \\
3818
+
\hline
3819
+
node\_id\_mapper\_file (filepath) & No default & & If using Node ID indexing, the file containing the node ID index list \\
3820
+
\hline
3821
+
random\_indexer\_seed (long) & System time & & The seed to use for a random allocation. If not specified, system time is used. \\
3822
+
\hline
3823
+
allocation (string) & first\_available & first\_available, random, cart, node\_id, coordinate & The scheme to use for allocating compute nodes to a given job. \\
3824
+
\hline
3825
+
random\_allocation\_seed (long) & System time & & For random allocation policy. If unspecified, system time is used as the seed. \\
3826
+
\hline
3827
+
node\_id\_allocation\_file (filepath) & No default & & If using Node ID allocation, the file containing the list of node IDs to allocate for the job \\
3828
+
\hline
3829
+
dumpi\_metaname (filepath) & No default & & If running DUMPI trace, the location of the metafile for configuring trace replay \\
3830
+
\hline
3831
+
coordinate\_file (filepath) & No default & & If running using coordinate allocation or indexing, the path to the file containing the node coordinates of each proc (MPI rank) \\
3832
+
\hline
3833
+
cart\_sizes (vector of int) & No default & & Launch a contiguous block of nodes in a Cartesian topology. This gives the size of each dimension in the block. \\
3834
+
\hline
3835
+
cart\_offsets (vector of int) & No default & & Launch a contiguous block nodes in a Cartesian topology. This gives the offset in each dimension where the block begins. \\
3836
+
\hline
3837
+
parsedumpi\_timescale (double) & 1.0 & Positive float & If running DUMPI traces, scale compute times by the given value. Values less than 1.0 speed up computation. Values greater than 1.0 slow down computation. \\
3838
+
\hline
3839
+
parsedumpi\_terminate\_percent (int) & 100 & 1-100 & Percent of trace. Can be used to terminate large traces early \\
3840
+
\hline
3841
+
host\_compute\_timer (bool) & False & & Use the compute time on the host to estimate compute delays \\
3842
+
\hline
3843
+
otf2\_metafile (string) & No default & string & The root file of an OTF2 trace. \\
3844
+
\hline
3845
+
otf2\_timescale (double) & 1.0 & Positive float & If running OTF2 traces, scale compute times by the given value. Values less than 1.0 speed up computation. Values greater than 1.0 slow down computation. \\
3846
+
\hline
3847
+
otf2\_print\_mpi\_calls (bool) & false & & Print MPI calls found in the OTF2 trace
3848
+
\\
3849
+
\hline
3850
+
otf2\_print\_trace\_events (bool) & false & & Debugging flag that printsindividual trace events (which includes details such as when an MPI call begins, ends, and when a collective begins and ends
3851
+
\\
3852
+
\hline
3853
+
otf2\_print\_time\_deltas (bool) & false & & Debugging flag that prints compute delays injected by the simulator
3854
+
\\
3855
+
\hline
3856
+
otf2\_warn\_unknown\_callback (bool) & false & & Debugging flag the prints unknown callbacks
0 commit comments