|
|
|
Making a list of nodes
|
|
|
|
----------------------
|
|
|
|
To launch the test, perform the learning phase and compute the coefficients, you will have to write a lists of nodes. A node list is just a text file with a hostname per line. For example
|
|
|
|
To launch the test, perform the learning phase and compute the coefficients, you will have to write a lists of nodes. A node list is just a text file with a hostname per line. For example:
|
|
|
|
|
|
|
|
```
|
|
|
|
node0001
|
| ... | ... | @@ -8,6 +8,13 @@ node0002 |
|
|
|
node0003
|
|
|
|
```
|
|
|
|
|
|
|
|
You can write 1 or N lists of nodes, depending on your plan or your cluster. For example, if the super computer have 5 islands of 500 nodes, you can perform the learning phase in just 40 nodes per island or in all nodes in all islands. The EAR team recommendation is to select a bunch of nodes of each of the total N island and write N lists of a small set of nodes.
|
|
|
|
Depending on your `ear.conf` configuration, having N number of islands, write N nodes lists with a bunch of nodes to perform the learning phase. It is not necessary to launch the learning phase in all nodes of your island, just a small percentage will be enough:
|
|
|
|
|
|
|
|
```
|
|
|
|
# ls
|
|
|
|
nodes.island1.txt
|
|
|
|
nodes.island2.txt
|
|
|
|
nodes.island3.txt
|
|
|
|
```
|
|
|
|
|
|
|
|
Continue the learning phase following the [next section](3-˗-Kernels-test). |
|
|
\ No newline at end of file |
