Lab 09 - Slurm practice

Learning goals

• Review of parallel computing
• Draw pretty networks with Slurm.

Review of parallel computing

Regardless of R–and in the context of an HPC cluster–we have various ways of parallelizing jobs:

1. At the core level, using SIMD instructions.

2. At the CPU level using multiple cores within a program (e.g., loop-level).

3. Same, but replicating the job within a node.

4. At the node level, replicating the job across nodes.

Task 1: Alternatives to job arrays

Besides job arrays in Slurm, look for an alternative way of replicating jobs across multiple nodes (hint: There’s a type of parallel::cluster object we haven’t discussed.) There are at least two.

Task 2: Your expert opinion

Your expert opinion has been requested. For each way to parallelize a job, provide: (i) a one-paragraph description, (ii) proposed criteria of when to use it, and (iii) an example of an analysis that could be done with it.

Draw pretty networks with Slurm

Draw a pretty picture of the yeast network using the development version of netplot:

library(netplot)
library(igraph)
library(igraphdata)

data(yeast)

set.seed(889)

# Netplot objects are grobs (like ggplot2)
np <- nplot(
    yeast,
    # Too many vertices, so just a sample
    sample.edges = .1 
    )

# So we need to print them
print(np)

Read the manual of nplot and try:

1. Coloring the vertices by the class Class attribute. You can access them using V(yeast)$Class.

2. Playing with the layout, see igraph’s ?layout_.

3. Tryout grid::patterns for the vertices, see what you can do.

Create a design

Task 1: Get the package

Install the package using git clone + R CMD INSTALL in the command line.

Task 2: First run

Try drawing one network with it. Use CHPC’s ondemand.

Task 3: Submit to CHPC

Write an R and Slurm script to submit the job to CHPC. The job should result in saving the netplot object to an RDS file and a png figure of 1024 x 780 px resolution.