How to disable Intel Hyper Threading Technology

Since ParallelCluster 2.5.0 you can disable hyperthreding by configuring the disable_hyperthreading parameter in the cluster section.

If you're using ParallelCluster 2.4.1 or older versions there are two things you need to do disable hyperthreading (on Amazon Linux).

1. Set scheduler to use cores

You have to pass an additional setting at cluster creation time, by using the extra_json parameter, which enable you to set the number of slots per instances to vpcus, cores, or a number.

To set the number of slots per instance to be only the number of cores, you have to set the following in your config:

[cluster yourcluster]
...
extra_json = { "cluster" : { "cfn_scheduler_slots" : "cores" } }

2. Turn off hyper-threading in a post-install script

If you are using Amazon Linux, you have to create a script with the following content:

#!/bin/bash

for cpunum in $(cat /sys/devices/system/cpu/cpu*/topology/thread_siblings_list | cut -s -d, -f2- | tr ',' '\n' | sort -un)
do
    echo 0 > /sys/devices/system/cpu/cpu$cpunum/online
done

Upload the script with the correct permissions to S3 and use as Post Install script by configuring the post_install configuration parameter.

[cluster yourcluster]
...
post_install = s3://<bucket-name>/yourscript.sh

See Disabling Intel Hyper-Threading Technology on Amazon Linux for more details.