Re: [RFC][PATCH 2/4] tracing: Use pid bitmap instead of a pid array for set_event_pid

[Mathieu Desnoyers]

----- On Apr 19, 2016, at 3:41 PM, rostedt rostedt@xxxxxxxxxxx wrote:

> On Tue, 19 Apr 2016 11:57:32 -0700
> "H. Peter Anvin" <hpa@xxxxxxxxx> wrote:
> 
>> Also, I understand there is one of these bitmaps per ring buffer, and
>> the ring buffer is in the tens of megabytes.
> 
> Right, there's only one bitmap per tracing instance, which in most
> cases is just one (I know of people who make more). And by default, the
> tracing buffer is 1.4 megs per CPU.
> 
> If you have a pid_max of the max size, I highly doubt you will be doing
> that on a single CPU machine. If you have 48 CPUs, the ring buffer will
> be 1.4 * 48 megs, making the 1/2 meg bitmap a nit.
> 
> I will say, there may be two bitmaps soon, because I plan on adding
> this same code to the function tracer logic.

Ah indeed, since there is a hard limit to 4194304, that makes the
worse case bitmap 512k.

We could argue that given a sparse dataset in the PID table (typical
in our use-cases), a small hash table would have better cache locality
than the bitmap. But I agree that the hash table does add a bit of
complexity, so it becomes a complexity vs cache locality tradeoff.
So I understand why you would want to go for the simpler bitmap
solution, unless the hash table would prove to bring a measurable
performance improvement.

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com