Re: lttng-dev Digest, Vol 203, Issue 7

[Olivier Dion via lttng-dev <lttng-dev@lists.lttng.org>]

On Fri, 21 Mar 2025, Kienan Stewart via lttng-dev <lttng-dev@lists.lttng.org> wrote:
> Hi Lakshya,
>
> On 3/21/25 8:52 AM, Gour DEV wrote:
>> Thanks Kienan for digging this and explaining the behaviour,
[...]
>> Also, I thought using --buffers-global
>> <https://lttng.org/man/1/lttng-enable-channel/v2.8/#doc-opt--buffers-global>
>> would
>> be better for me as I could know how much memory will be required but looks
>> like that option is only available for kernel tracking, it would be very
>> good if this option could be available for user buffering also, but that
>> could come later.
>> 
>
> Global buffers for UST is planned for the next release. The trade-off of 
> using it is that tracing overhead will increase (as the buffers are no 
> longer per-CPU), but in exchange for a potentially lower memory 
> footprint particularly in situations where not all possible CPUs are online.
>

Little correction here.  `--buffers-global' is the ownership of the
buffers (UID vs PID vs system).  The next release introduces the buffers
allocation policies `per-{cpu,channel}'.  Therefore `--buffers-global'
will still be only available for the kernel domain, AFAIK.

User-space however will be able to allocate a single ring-buffer for a
given channel (actually it depends on the buffer ownership UID vs PID),
thus reducing the amount of memory used, for that channel, by a factor
proportional to the number of CPUs possible on the system.

[...]
 
Thanks,
Olivier
-- 
Olivier Dion
EfficiOS Inc.
https://www.efficios.com