Re: rte_ring features in use (or not)

On Wed, Jan 25, 2017 at 03:59:55PM +0000, Wiles, Keith wrote:

Sent from my iPhone

quoted

On Jan 25, 2017, at 7:48 AM, Bruce Richardson [off-list ref] wrote:

quoted

On Wed, Jan 25, 2017 at 01:54:04PM +0000, Bruce Richardson wrote:

quoted

On Wed, Jan 25, 2017 at 02:20:52PM +0100, Olivier MATZ wrote:
On Wed, 25 Jan 2017 12:14:56 +0000, Bruce Richardson
[off-list ref] wrote:

quoted

Hi all,

while looking at the rte_ring code, I'm wondering if we can simplify
that a bit by removing some of the code it in that may not be used.
Specifically:

* Does anyone use the NIC stats functionality for debugging? I've
 certainly never seen it used, and it's presence makes the rest less
 readable. Can it be dropped?

What do you call NIC stats? The stats that are enabled with
RTE_LIBRTE_RING_DEBUG?

Yes. By NIC I meant ring. :-(

quoted

<snip>

quoted

quoted

For the ring, in my opinion, the stats could be fully removed.

That is my thinking too. For mempool, I'd wait to see the potential
performance hits before deciding whether or not to enable by default.
Having them run-time enabled may also be an option too - if the branches
get predicted properly, there should be little to no impact as we avoid
all the writes to the stats, which is likely to be where the biggest hit
is.

quoted

quoted

* RTE_RING_PAUSE_REP_COUNT is set to be disabled at build time, and
 so does anyone actually use this? Can it be dropped?

This option looks like a hack to use the ring in conditions where it
should no be used (preemptable threads). And having a compile-time
option for this kind of stuff is not in vogue ;)

<snip>

quoted

quoted

quoted

* Who uses the watermarks feature as is? I know we have a sample app
 that uses it, but there are better ways I think to achieve the same
 goal while simplifying the ring implementation. Rather than have a
set watermark on enqueue, have both enqueue and dequeue functions
return the number of free or used slots available in the ring (in
case of enqueue, how many free there are, in case of dequeue, how
many items are available). Easier to implement and far more useful to
the app.

+1

Bonus question:
* Do we know how widely used the enq_bulk/deq_bulk functions are? They
 are useful for unit tests, so they do have uses, but I think it would
 be good if we harmonized the return values between bulk and burst
 functions. Right now:
   enq_bulk  - only enqueues all elements or none. Returns 0 for all, or
               negative error for none.
   enq_burst - enqueues as many elements as possible. Returns the number
               enqueued.

I do use the apis in pktgen and the difference in return values has got me once. Making them common would be great,  but the problem is backward compat to old versions I would need to have an ifdef in pktgen now. So it seems like we moved the problem to the application.

Yes, an ifdef would be needed, but how many versions of DPDK back do you
support? Could the ifdef be removed again after say, 6 months?

I would like to see the old API kept and a new API with the new behavior. I know it adds another API but one of the API would be nothing more than wrapper function if not a macro. 

Would that be more reasonable then changing the ABI?

Technically, this would be an API rather than ABI change, since the
functions are inlined in the code. However, it's not the only API change
I'm looking to make here - I'd like to have all the functions start
returning details of the state of the ring, rather than have the
watermarks facility. If we add all new functions for this and keep the
old ones around, we are just increasing our maintenance burden.

I'd like other opinions here. Do we see increasing the API surface as
the best solution, or are we ok to change the APIs of a key library like
the rings one?

/Bruce