Babeltrace 2.0.2 performance issue

Babeltrace 2.0.2 performance issue

[Aleksander Aleksandrov via lttng-dev]

[-- Attachment #1.1: Type: text/plain, Size: 1277 bytes --]

От: Aleksander Aleksandrov

Отправлено: 12 марта 2020 г. 14:36
Кому: jeremie.galarneau@efficios.com
Копия: simon.marchi@efficios.com; francis.deslauriers@efficios.com
Тема: Babeltrace 2.0.2 performance issue


Hello Jeremie,


My name is Aleksandr, I had a pleasure to write a plugin for babeltrace2 and I faced some performance issue (bt2 vs bt1).

I attached the simple LTTng user-space log which is used to show my issue. As far as I understand Babeltrace2's performance is comparable to Babeltrace1's.

  *   Babeltrace 2:

tag: v2.0.2

BABELTRACE_MINIMAL_LOG_LEVEL=INFO ./configure --disable-man-pages --disable-debug-info CFLAGS='-O3 -DNDEBUG -Wno-error'

LD_LIBRARY_PATH=./src/lib/.libs ./src/cli/.libs/babeltrace2 my-lttng-trace -o dummy  12,70s user 0,03s system 99% cpu 12,745 total

  *   Babeltrace 1:

tag: v1.5.8

./configure CFLAGS='-O3 -DNDEBUG -Wno-error'

./converter/babeltrace my-lttng-trace -o dummy  8,30s user 0,03s system 100% cpu 8,334 total


Babeltrace v2.0.2 is slower (~50%, processing time: 12,745 s vs 8,334 s) than Babeltrace v1.5.8 with a dummy output according to my measurements.


I would be happy to receive any help or clarification from you.

Thank you for your time!


Best regards,

Aleksandr

[-- Attachment #1.2: Type: text/html, Size: 4780 bytes --]

[-- Attachment #2: Type: text/plain, Size: 156 bytes --]

_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

Babeltrace 2.0.2 performance issue

[Aleksander Aleksandrov via lttng-dev]

[-- Attachment #1.1: Type: text/plain, Size: 1219 bytes --]

Hi Simon,


Thank you for your message! Unfortunately, I missed your previous message, sorry for this.


I processed the CI's trace logs, I got the following performance: bt2 (140.51s) vs bt1 (122.00s). In my opinion, the values are quite similar comparing with my other measurements.


Best regards,

Aleksandr


On 2020-03-13 4:58 p.m., Simon Marchi wrote:
> Hi Aleksander,
>
> I just noticed you did not send your original email to the lttng-dev mailing list,
> please send such request on that mailing list, as it's of public interest:
>
>   https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
>
> See my previous response here:
>
>   https://pastebin.com/raw/3Q5PbYXn
>
> Simon
>

Hi Aleksander,

Sorry, I forgot to follow up.  Did you get my last message?

This is the trace we use for benchmarking: https://files.efficios.com/s/pog5raGkBkH63y9

Could you try to compare bt1 and bt2 using that trace?

On the CI benchmark system, we get some similar performance with both bt1 and bt2.  Although
a colleague tried on a recent AMD Ryzen 3700X CPU, and he says bt2 is slower than bt1 for
him (36s vs 33s).  I'd be curious to know what kind of numbers you get.

Simon

[-- Attachment #1.2: Type: text/html, Size: 3791 bytes --]

[-- Attachment #2: Type: text/plain, Size: 156 bytes --]

_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

Re: Babeltrace 2.0.2 performance issue

[Jonathan Rajotte-Julien via lttng-dev]

Hi Aleksander,

On Thu, Apr 02, 2020 at 04:32:03PM +0000, Aleksander Aleksandrov via lttng-dev wrote:
> Hi Simon,
> 
> 
> Thank you for your message! Unfortunately, I missed your previous message, sorry for this.
> 
> 
> I processed the CI's trace logs, I got the following performance: bt2 (140.51s) vs bt1 (122.00s). In my opinion, the values are quite similar comparing with my other measurements.

Well, in that case we will augment the data gathered on our CI to better
understand what is happening here an why we have such discrepancy. I should be
able to allocate some time in the following weeks.

Feel free to contribute to the optimization effort on your end if you see any
big culprit. For now, the babeltrace team effort is primary on completing
documentation and internal cleanup.

On my laptop (Intel(R) Core(TM) i7-7500U CPU @ 2.70GHz) with scaling enabled
(~3.2Ghz mean of freq used during the 5 run of each bt1 and bt2) I get extremely
similar result for both bt1 (~46s) and bt2(~45s).

On my desktop as Simon reported, (Ryzen 7 3700x , max freq 3.6Ghz), bt1 is ~33s and bt2 ~36s.

All of this using the dummy output for both bt1 and bt2.

Cheers

> 
> 
> Best regards,
> 
> Aleksandr
> 
> 
> On 2020-03-13 4:58 p.m., Simon Marchi wrote:
> > Hi Aleksander,
> >
> > I just noticed you did not send your original email to the lttng-dev mailing list,
> > please send such request on that mailing list, as it's of public interest:
> >
> >   https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
> >
> > See my previous response here:
> >
> >   https://pastebin.com/raw/3Q5PbYXn
> >
> > Simon
> >
> 
> Hi Aleksander,
> 
> Sorry, I forgot to follow up.  Did you get my last message?
> 
> This is the trace we use for benchmarking: https://files.efficios.com/s/pog5raGkBkH63y9
> 
> Could you try to compare bt1 and bt2 using that trace?
> 
> On the CI benchmark system, we get some similar performance with both bt1 and bt2.  Although
> a colleague tried on a recent AMD Ryzen 3700X CPU, and he says bt2 is slower than bt1 for
> him (36s vs 33s).  I'd be curious to know what kind of numbers you get.
> 
> Simon

> _______________________________________________
> lttng-dev mailing list
> lttng-dev@lists.lttng.org
> https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev


-- 
Jonathan Rajotte-Julien
EfficiOS