Re: [PATCHSET 00/16] perf top: Add multi-thread support (v1)
From: Namhyung Kim
Date: Mon Dec 14 2015 - 12:07:42 EST
On Mon, Dec 14, 2015 at 07:46:28AM -0700, David Ahern wrote:
> On 12/14/15 2:26 AM, Peter Zijlstra wrote:
> >On Fri, Dec 11, 2015 at 08:01:31AM -0700, David Ahern wrote:
> >>On 12/11/15 1:11 AM, Ingo Molnar wrote:
> >>>* Namhyung Kim <namhyung@xxxxxxxxx> wrote:
> >>>>IIRC David said that thread per cpu seems too much especially on a large system
> >>>>(like ~1024 cpu). [...]
> >>>Too much in what fashion? For recording I think it's the fastest, most natural
> >>>model - anything else will create cache line bounces.
> >>The intrusiveness of perf on the system under observation. I understand
> >>there are a lot of factors that go into it.
> >So I can see some of that, if every cpu has its own thread then every
> >cpu will occasionally schedule that thread. Whereas if there were less,
> >you'd not have that.
> >Still, I think it makes sense to implement it, we need the multi-file
> >option anyway. Once we have that, we can also implement a per-node
> >option, which should be a fairly simple hybrid of the two approaches.
> >The thing is, perf-record is really struggling on big machines.
> I've gone from the 1024-cpu sparc systems earlier this year down to small
> PPC and Rangeley-based switches. For both ends of the scale (and in between)
> I constantly struggle with the options to manage memory, cpu and disk
> There definitely needs to be options (e.g., multi-threaded on/off). For the
> threading options I get the appeal for 1-thread per cpu but other options
> make sense as well -- 1 thread per core, 1 per NUMA node. perf has the CPU
> topology so should not be too difficult.
I think we can use --num-thread option to control multi-threading: 1
for disabling and others for enabling. In the current implementation,
using 1 thread still use same logic so 1 reader + 1 collector will be
created as well as 1 display thread. Not sure it'd be better special
casing 1 thread to use different code path.
Anyway, I think it'd be nice to have per-core, per-socket and per-node
options and per-core is a good default then.
> If you have 1-thread per cpu that means you are pinning the threads to the
> cpu? That brings in additional permissions problems.
Did you mean setting sched affinity? It seems not a privileged
operation doing it for its own threads..
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/