[lkml]   [2009]   [Nov]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Subject[PATCH 0/2] [GIT PULL] tracing: performance boosts for recording traces


The first patch is really a fix to the ring_buffer_benchmark. It was
not giving a correct representation of the time of a trace. The
logic to time the trace was 1/3 of the measurement.

The second patch removes the disabling of interrupts in trace_clock_local.
This showed a 30 to 40 ns latency.

I'm still seeing the time stamp being a large overhead in the tracing.
I'll be posting an RFC patch set on using normalizing the timestamp
on read, to take some of the work out of the write side. This
drops the write from 179 ns to 151 ns. More about that later.

Please pull the latest tip/tracing/core tree, which can be found at:


Steven Rostedt (2):
ring-buffer: Add multiple iterations between benchmark timestamps
tracing: do not disable interrupts for trace_clock_local

kernel/trace/ring_buffer_benchmark.c | 25 ++++++++++++++++---------
kernel/trace/trace_clock.c | 8 +++++---
2 files changed, 21 insertions(+), 12 deletions(-)

 \ /
  Last update: 2009-11-12 05:53    [from the cache]
©2003-2014 Jasper Spaans. hosted at Digital Ocean