How to provide hugepages configuration

[eroussy]

Abstract
Two different mechanism are used to add hugepages:

• Directly add it in the kernel command line
• Use the service hugetlb-gigantic-pages.service
  We should use and test only one mechanism.

Current status
On Yocto :

• Kernel parameters default_hugepagesz=1G hugepagesz=1G are added by default.
  Method 1 :
• Add N hugepages by adding parameter hugepages=N to this command line
  Method 2 :
• Ansible variable yocto_hugepages
• Uses roles roles/yocto/hugepages
• Triggers service hugetlb-gigantic-pages.service on the hypervisor

On Debian:

• Nothing is provided nor documented
• Only solution is to add kernel parameters default_hugepagesz=1G hugepagesz=1G hugepages=N
• hugetlb-gigantic-pages.service was originally provided, but removed in 0e9b0ab

Detailed Description
The advantage of the hugetlb-gigantic-pages.service is that hugepages can be changed at runtime. However, the kernel can fail to allocate these hugepages if the RAM is already too much used.
I think the simplest solution is to do the hugepages allocation only through command line, for both Yocto and Debian.

Benefits to SEAPATH

• Simpler for new user
• Clearer
• Same solutions for both Yocto and Debian

Work to be done

• Prerequisites : Harmonize kernel parameters Harmonize kernel parameters add #730
• Remove hugetlb-gigantic-pages.service from yocto build in https://github.com/seapath/meta-seapath
• Remove default kernel parameters default_hugepagesz=1G hugepagesz=1G in https://github.com/seapath/meta-seapath
• Document how to add hugepages
• update ABB inventory example

Note: once this task is done, we shouln't need the roles/yocto directory anymore.

[eroussy]

@insatomcat @dupremathieu What do you think of this proposition ?