Re: [[RFC][PATCH] v1 0/2] malloc/realloc with transparent huge page support

[Norbert Manthey <nmanthey@conp-solutions.com>]

On 04.05.20 06:36, Florian Weimer wrote:
> * Norbert Manthey:
>
>> This change has been tested on top of glibc 2.23 in an Ubuntu 16.04
>> environment for programs that benefit from huge pages due to their
>> large memory usage and pseudo-random memory access patterns
>> (e.g. SAT solvers, model checkers, optimization tools an
>> others). More details on the performance improvements for these
>> tools can be found in https://arxiv.org/abs/2004.14378, e.g. page 9.
> Please show us the output of
>
> $ cat /sys/kernel/mm/transparent_hugepage/enabled
>
> from your test systems.  Thanks.

I have seen [madvise] as the default for different Ubuntu versions
(16.04, 18.04 and 20.04), so I expect it to be around for at least the
next 5 years.

$ cat /sys/kernel/mm/transparent_hugepage/enabled
always [madvise] never

>
> In the paper, you say this:
>
> | Because there might be applications running on the host that would
> | suffer from larger pages, THP is usually disabled on physical
> | systems and it is not advised to set the value to always.
>
> However, the default value for Red Hat Enterprise Linux 7 is, in fact,
> "always", so it is puzzling why you see such a large benefit for
> long-running processes.

All used test systems have not been using [always], but [madvise]. For
systems that already use [always], I expect the benefit to be much
smaller. With the 2M alignment, there should be less 4K pages required,
but we did not really measure the effect of only using the alignment.

Best,
Norbert