RE: [RFC PATCH v11] pile stack and mempool driver

["Morten Brørup" <mb@smartsharesystems.com>]

> From: Stephen Hemminger [mailto:stephen@networkplumber.org]
> Sent: Tuesday, 11 August 2026 17.25
> 
> On Tue, 11 Aug 2026 12:35:41 +0000
> Morten Brørup <mb@smartsharesystems.com> wrote:
> 
> > Early submission of:
> > - A new "pile" stack-like implementation using the Stack API, and
> > - an accompanying "pile" mempool driver.
> >
> > For CI test and community feedback.
> >
> > Needless to say, this must be separated into multiple independent
> > series of patches.
> > And release notes must be added.
> > For now, I'm submitting a snapshot of work in progress.
> >
> > The "pile" somewhat resembles the lock-free stack, but operates on
> > bulks (arrays) of objects, to significantly reduce linked list
> > traversal.
> > With the pile's default bulk size of 32 objects, a mempool cache
> > flush/refill traverses a linked list of only 16 elements, whereas
> > the lock-free stack would traverse a linked list of 512 elements.
> >
> > Some performance numbers from mempool_perf_autotest_2cores, all
> > with cache=1024 cores=2 n_keep=32768:
> >
> > start performance test (using ring_mp_mc, with cache)
> > n_get_bulk= 64 n_put_bulk= 64 constant_n=0 rate_persec= 753985338
> > n_get_bulk=256 n_put_bulk=256 constant_n=0 rate_persec= 755805913
> >
> > start performance test for lf_stack (with cache)
> > n_get_bulk= 64 n_put_bulk= 64 constant_n=0 rate_persec=  29132352
> > n_get_bulk=256 n_put_bulk=256 constant_n=0 rate_persec=  29276708
> >
> > start performance test for pile (with cache)
> > n_get_bulk= 64 n_put_bulk= 64 constant_n=0 rate_persec= 560159479
> > n_get_bulk=256 n_put_bulk=256 constant_n=0 rate_persec= 557910933
> >
> > Hat tip to Bruce for bringing attention to the ring not being the
> > optimal mempool driver!
> >
> > Note: The GitHub "mini" tests don't include the "pile" mempool
> driver,
> > and are expected to fail.
> >
> > Signed-off-by: Morten Brørup <mb@smartsharesystems.com>
> 
> Right but the 32 bit code fails at run time.
> 
>  MEMPOOL: Cache size 250 not divisible by 32, rounding down to 224.
> STACK: rte_stack_create(): Pile is not supported on your platform

The "pile" and "lf_stack" stack types (and mempool drivers) rely on a 128-bit CAS instruction, which is only available on x86_64 and arm64, so asking for one of those mempool drivers will fail on 32-bit architectures.

> 
> Should be possible to correct this at compile. Like if pile is not
> available on this platform, something else should be the default.

Yep. The default will remain the "ring" mempool driver.
I only temporarily selected the "pile" driver as the default to ensure that it got properly tested by the CI.

> 
> The CI failures indicate places where it is likely to be a problem
> for users.

Other CI failures are related to "mini" builds.
The "mini" builds only include the "ring" mempool driver, so selecting the "pile" mempool driver naturally fails:
MEMPOOL: Unknown mempool_ops <pile>, of 6 ops registered

I think v11 has brought sufficient CI/AI feedback to split it into individual patch series for the "pile" stack/driver and the mempool optimizations.
Will proceed with that now.

Thank you for the advanced AI reviews, Stephen; they were helpful.

-Morten