[Bug runtime/9822] New: reduce timer load

[Bug runtime/9822] New: reduce timer load

[fche at redhat dot com]

Two sources of timer/wakeup traffic from the systemtap runtime
should be reduced:

- the runtime/time.c timekeeping widgetry that uses jiffies to keep the per-cpu
khz estimates updated

- the runtime/transport/utt.c stuff that flushes buffers to userspace

-- 
           Summary: reduce timer load
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: runtime
        AssignedTo: systemtap at sources dot redhat dot com
        ReportedBy: fche at redhat dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug runtime/9822] reduce timer load

[wcohen at redhat dot com]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |wcohen at redhat dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug runtime/9822] reduce timer load

[mhiramat at redhat dot com]

------- Additional Comments From mhiramat at redhat dot com  2009-02-04 23:41 -------
(In reply to comment #0)
> - the runtime/time.c timekeeping widgetry that uses jiffies to keep the per-cpu
> khz estimates updated

This ensures the accuracy of the self-own version of gettimeofday_*.
- if the script doesn't use gettimeofday_*, just disable timer. And if the
script use gettimeofday_s, increase the timer interval.
or
- after cpu-khz estimating period, don't use timer(or increase interval to 1sec)
until cpu-freq is changed.
or 
- improve upstream kernel not to use locks.

> - the runtime/transport/utt.c stuff that flushes buffers to userspace

This is for avoiding deadlock of global timer lock. I think one possible
solution is using tracepoint in scheduler to wakeup waiting reader thread as
ftrace do.


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug runtime/9822] reduce timer load

[ananth at in dot ibm dot com]

------- Additional Comments From ananth at in dot ibm dot com  2009-02-20 06:49 -------
Any updates on this? From what I hear, this is limiting SystemTap's usefulness
in work that involves power management.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ananth at in dot ibm dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug runtime/9822] reduce timer load

[fche at redhat dot com]

------- Additional Comments From fche at redhat dot com  2009-02-20 12:39 -------
> Any updates on this?

No one has yet grabbed this bug.


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug runtime/9822] reduce timer load

[mhiramat at redhat dot com]

------- Additional Comments From mhiramat at redhat dot com  2009-02-23 20:52 -------
(In reply to comment #1)
> > - the runtime/transport/utt.c stuff that flushes buffers to userspace
> 
> This is for avoiding deadlock of global timer lock. I think one possible
> solution is using tracepoint in scheduler to wakeup waiting reader thread as
> ftrace do.

Hmm, as far as I can see, current ftrace has deadlock problem...




-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug runtime/9822] reduce timer load

[jistone at redhat dot com]

------- Additional Comments From jistone at redhat dot com  2009-02-23 23:29 -------
(In reply to comment #1)
> > - the runtime/time.c timekeeping widgetry that uses jiffies to keep the
> > per-cpu khz estimates updated
> 
> This ensures the accuracy of the self-own version of gettimeofday_*.
> - if the script doesn't use gettimeofday_*, just disable timer.

I just committed a change to this effect, so we should be in a lot better shape.
 For an idle script that doesn't call gettimeofday, powertop tells me that I
went from 88.9% C3 residency to 97.1% (vs. 98.3% without stap).

Many of the other suggestions here are still valid, but I think this was the
biggest step.

-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9822

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.