[lttng-dev] UST performance

Mathieu Desnoyers mathieu.desnoyers at efficios.com
Tue Apr 28 11:47:38 EDT 2015


Is your sched_getcpu() system call a vdso too ? This may explain the difference. 

Moreover, make sure you #define _LGPL_SOURCE in your benchmark program 
before including any lttng ust header. It will inline many functions associated with 

Then, you can try using "perf" to profile the runtime execution of your program 
traced by lttng-ust ? This should give us the information we need to investigate the 
performance profile. 



----- Original Message -----

> Hi,dev

> About a month ago,I asked a question about how to do time recording for ust
> event when vdso_clock_gettime not availble on my platform(Glibc too
> old).Recently I implemented a lib to analyze vdso in memory and lookuped the
> symbol vdso_clock_gettime successfully.With this vdso_clock_gettime
> availble,I get a 16% improvement in time spending per ust event. I.E. I
> measured the time taked by a event with 10 bytes string :

> no vdso_clock_gettime : 0.642us
> with vdso_clock_gettime:0.554us

> However,Both the numbers are much higher than the following 0.28us(But closed
> to 0.5us --W/O opt). So what's the difference between UST(0.28us) and UST
> W/O OPT(0.5us)??

> PS:My hardware
> 2 x Xeon E5-2430 2.20GHz 100MHz FSB (24 cores) 94.6GB / 96GB 1333MHz DDR3 ==
> 6 x 16GB, 6 x empty

> Thanks
> zhenyu.ren

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

Mathieu Desnoyers 
EfficiOS Inc. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lttng.org/pipermail/lttng-dev/attachments/20150428/19a46d40/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: temp4cj.png
Type: image/png
Size: 34762 bytes
Desc: not available
URL: <http://lists.lttng.org/pipermail/lttng-dev/attachments/20150428/19a46d40/attachment-0001.png>

More information about the lttng-dev mailing list