lkml.org 
[lkml]   [2010]   [Feb]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[RFC GIT PULL] perf/trace/lock optimization/scalability improvements
Date
Hi,

There are many things that happen in this patchset, treating
different problems:

- remove most of the string copy overhead in fast path
- open the way for lock class oriented profiling (as
opposite to lock instance profiling. Both can be useful
in different ways).
- remove the buffers muliplexing (less contention)
- event injection support
- remove violent lock events recursion (only 2 among 3, the remaining
one is detailed below).

Some differences, by running:
perf lock record perf sched pipe -l 100000

Before the patchset:

Total time: 91.015 [sec]

910.157300 usecs/op
1098 ops/sec

After this patchset applied:

Total time: 43.706 [sec]

437.062080 usecs/op
2288 ops/sec

Although it's actually 50 secs after the very latest patch in this
series. It is supposed to bring more scalability (and I believe it
does on a box with more than two cpus, although I can't test).
But multiplexing the counters had a side effect: perf record has
only one buffer to eat and not 5 * NR_CPUS, which makes its job
a bit easier when we multiplex (at the cost of cpus contention of
course, but on my atom, the scalability gain is not very visible).

And also, after this odd patch:

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 98fd360..254b3d4 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -3094,7 +3094,8 @@ static u32 perf_event_tid(struct perf_event *event, struct task_struct *p)
if (event->parent)
event = event->parent;

- return task_pid_nr_ns(p, event->ns);
+ return p->pid;
}

We get:

Total time: 26.170 [sec]
261.707960 usecs/op
3821 ops/sec

Ie: 2x faster than this patchset, and more than 3x faster than
tip:/perf/core

This is because task_pid_nr_ns() takes a lock and creates
lock events recursion. We really need to fix that.

You can pull this patchset from:

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
perf/core

Thanks.


---
Frederic Weisbecker (11):
tracing: Add lock_class_init event
tracing: Introduce TRACE_EVENT_INJECT
tracing: Inject lock_class_init events on registration
tracing: Add lock class id in lock_acquire event
perf: New PERF_EVENT_IOC_INJECT ioctl
perf: Handle injection ioctl with trace events
perf: Handle injection iotcl for tracepoints from perf record
perf/lock: Add support for lock_class_init events
tracing: Remove the lock name from most lock events
tracing/perf: Fix lock events recursions in the fast path
perf lock: Drop the buffers multiplexing dependency


include/linux/ftrace_event.h | 6 +-
include/linux/lockdep.h | 4 +
include/linux/perf_event.h | 6 +
include/linux/tracepoint.h | 3 +
include/trace/define_trace.h | 6 +
include/trace/events/lock.h | 57 ++++--
include/trace/ftrace.h | 31 +++-
kernel/lockdep.c | 16 ++
kernel/perf_event.c | 47 ++++-
kernel/trace/trace_event_profile.c | 46 +++--
kernel/trace/trace_events.c | 3 +
tools/perf/builtin-lock.c | 345 ++++++++++++++++++++++++++++++++----
tools/perf/builtin-record.c | 9 +
13 files changed, 497 insertions(+), 82 deletions(-)

\
 
 \ /
  Last update: 2010-02-03 10:17    [W:0.175 / U:0.472 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site