f



futex high sys cpu utilization

Hi,=20

In my c++ (Linux) application I am seeing high CPU utilization (almost 90%)=
 and it that sys-cpu:user-cpu ratio is 8:2. I then tried to run =93strace=
=94 command and found that around 80% of time is being spent in =93futex=94=
 system call, please see snippet of strace here:=20

% time     seconds  usecs/call     calls    errors syscall=20
------ ----------- ----------- --------- --------- ----------------=20
 80.66  100.007743        1567     63806     22761 futex=20
  9.15   11.340274       96925       117         4 restart_syscall=20
  5.60    6.939598        4401      1577           poll=20

Initially I though this is because of pthread_lock, unlock, condition timed=
wait, condition wait, etc. system calls I am using in the application, to p=
rove that I have overridden pthread_lock, unlock, condition timedwait, cond=
ition wait, unlock, etc. using LD_PRELOAD functionality and concluded that =
not many threads are waiting here (mutex->__data.__nusers remains less than=
 5 most of the time), so it doesn=92t seem that these sys calls are part of=
 the problem.=20

Now I am running out of clue about how to find out code location/system cal=
l which is causing this futex high utilization, any help on this would be a=
ppreciated.=20

Thank you,=20
Jaydeep
0
7/17/2012 5:20:20 AM
comp.unix.programmer 10848 articles. 0 followers. kokososo56 (349) is leader. Post Follow

10 Replies
1394 Views

Similar Articles

[PageSpeed] 16

Jaydeep Chovatia , dans le message
<f87be638-dccd-4a9d-a092-ef9ae92e0290@googlegroups.com>, a écrit :
> In my c++ (Linux) application I am seeing high CPU utilization (almost
> 90%) and it that sys-cpu:user-cpu ratio is 8:2. I then tried to run
> “strace” command and found that around 80% of time is being spent in
> “futex” system call, please see snippet of strace here: 

If your uptime is more than 17 days, try:

sudo date -s "$(date)"

(or equivalent) and see if it fixes things. This is not a joke: there was a
leap second at the end of last month, and a strange bug in the NTP handling
causes futexes to expire prematurely. This exact command prevented Iceweasel
from eating all my CPU.
0
george54 (194)
7/17/2012 8:03:29 AM
Jaydeep Chovatia <chovatia.jaydeep@gmail.com> writes:
> In my c++ (Linux) application I am seeing high CPU utilization
> (almost 90%) and it that sys-cpu:user-cpu ratio is 8:2. I then tried
> to run “strace” command and found that around 80% of time is being
> spent in “futex” system call, please see snippet of strace here:
>
> % time     seconds  usecs/call     calls    errors syscall 
> ------ ----------- ----------- --------- --------- ---------------- 
>  80.66  100.007743        1567     63806     22761 futex 
>   9.15   11.340274       96925       117         4 restart_syscall 
>   5.60    6.939598        4401      1577           poll 
>
> Initially I though this is because of pthread_lock, unlock,
> condition timedwait, condition wait, etc. system calls I am using in
> the application, to prove that I have overridden pthread_lock,
> unlock, condition timedwait, condition wait, unlock, etc. using
> LD_PRELOAD functionality and concluded that not many threads are
> waiting here (mutex->__data.__nusers remains less than 5 most of the
> time), so it doesn’t seem that these sys calls are part of the
> problem.

Wild guess: Try running date -s "`date`". If this helps, you are
probably the last person on earth to notice that the Linux leap second
handling code is/was broken ...
0
rweikusat (2830)
7/17/2012 10:55:41 AM
On Tuesday, 17 July 2012 03:55:41 UTC-7, Rainer Weikusat  wrote:
> Jaydeep Chovatia &lt;chovatia.jaydeep@gmail.com&gt; writes:
> &gt; In my c++ (Linux) application I am seeing high CPU utilization
> &gt; (almost 90%) and it that sys-cpu:user-cpu ratio is 8:2. I then tried
> &gt; to run =93strace=94 command and found that around 80% of time is bei=
ng
> &gt; spent in =93futex=94 system call, please see snippet of strace here:
> &gt;
> &gt; % time     seconds  usecs/call     calls    errors syscall=20
> &gt; ------ ----------- ----------- --------- --------- ----------------=
=20
> &gt;  80.66  100.007743        1567     63806     22761 futex=20
> &gt;   9.15   11.340274       96925       117         4 restart_syscall=
=20
> &gt;   5.60    6.939598        4401      1577           poll=20
> &gt;
> &gt; Initially I though this is because of pthread_lock, unlock,
> &gt; condition timedwait, condition wait, etc. system calls I am using in
> &gt; the application, to prove that I have overridden pthread_lock,
> &gt; unlock, condition timedwait, condition wait, unlock, etc. using
> &gt; LD_PRELOAD functionality and concluded that not many threads are
> &gt; waiting here (mutex-&gt;__data.__nusers remains less than 5 most of =
the
> &gt; time), so it doesn=92t seem that these sys calls are part of the
> &gt; problem.
>=20
> Wild guess: Try running date -s &quot;`date`&quot;. If this helps, you ar=
e
> probably the last person on earth to notice that the Linux leap second
> handling code is/was broken ...

Thanks for your prompt response.
I had already tried the leap second solution but after that also i see this=
 problem.

Thank you,
Jaydeep
0
7/17/2012 3:46:58 PM
Nicolas George <nicolas$george@salle-s.org> wrote:
> Jaydeep Chovatia , dans le message
> <f87be638-dccd-4a9d-a092-ef9ae92e0290@googlegroups.com>, a écrit :
> > In my c++ (Linux) application I am seeing high CPU utilization (almost
> > 90%) and it that sys-cpu:user-cpu ratio is 8:2. I then tried to run
> > “strace” command and found that around 80% of time is being spent in
> > “futex” system call, please see snippet of strace here: 

> If your uptime is more than 17 days, try:

> sudo date -s "$(date)"

> (or equivalent) and see if it fixes things. This is not a joke: there was a
> leap second at the end of last month, and a strange bug in the NTP handling
> causes futexes to expire prematurely. This exact command prevented Iceweasel
> from eating all my CPU.

IIRC, the bug only causes this particular behavior when using sub-second
timeouts on a futex. The OP might also want to reconsider why he's using
sub-second timeouts in a loop.
0
William
7/17/2012 6:09:51 PM
William Ahern <william@wilbur.25thandClement.com> writes:
> Nicolas George <nicolas$george@salle-s.org> wrote:
>> Jaydeep Chovatia , dans le message
>> <f87be638-dccd-4a9d-a092-ef9ae92e0290@googlegroups.com>, a écrit :
>> > In my c++ (Linux) application I am seeing high CPU utilization (almost
>> > 90%) and it that sys-cpu:user-cpu ratio is 8:2. I then tried to run
>> > “strace” command and found that around 80% of time is being spent in
>> > “futex” system call, please see snippet of strace here: 
>
>> If your uptime is more than 17 days, try:
>
>> sudo date -s "$(date)"
>
>> (or equivalent) and see if it fixes things. This is not a joke: there was a
>> leap second at the end of last month, and a strange bug in the NTP handling
>> causes futexes to expire prematurely. This exact command prevented Iceweasel
>> from eating all my CPU.
>
> IIRC, the bug only causes this particular behavior when using sub-second
> timeouts on a futex. The OP might also want to reconsider why he's using
> sub-second timeouts in a loop.

This affects anything using a subsecond timeout together with
CLOCK_REALTIME hrtimer and the reason is these hrtimers maintain a
redundant copy of 'the current time' which is not updated from the
leap second insertion code because Thomas Gleixner killed the call
that was supposed to do this in 2007
(746976a301ac9c9aa10d7d42454f8d6cdad8ff2b).

In case this is of interest to someone: I could offer a working patch
for that (used in production) for anything which predates the
6b43ae8a619d17c4935c3320d2ef9e92bdeed05d kernel/time/ntp.c commit
(Thu Mar 15 13:04:03 2012 -0700, meaning, it uses the hrtimer-based
leap-second handling before John Stultz introduced a locking order
problem into that by splitting a certain lock into two locks).

The patch itself is against 2.6.36.4. It my introduce blasphemous
whitespaces. It was tested with the 'John Stultz' test case program.

Index: linux-2-6/kernel/time/ntp.c
===================================================================
RCS file: /sysdata/cvs/linux-2-6/kernel/time/ntp.c,v
retrieving revision 1.6
retrieving revision 1.7
diff -u -r1.6 -r1.7
--- linux-2-6/kernel/time/ntp.c	19 Aug 2010 17:04:51 -0000	1.6
+++ linux-2-6/kernel/time/ntp.c	4 Jul 2012 15:35:33 -0000	1.7
@@ -184,10 +184,21 @@
  * day, the system clock is set back one second; if in leap-delete
  * state, the system clock is set ahead one second.
  */
+static void call_clock_was_set(unsigned long unused)
+{
+	(void)unused;
+	clock_was_set();
+}
+
 static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
 {
+	static struct timer_list
+		call_clock_was_set_timer = TIMER_INITIALIZER(call_clock_was_set, 0, 0);
+	
 	enum hrtimer_restart res = HRTIMER_NORESTART;
+	int leap;
 
+	leap = 0;
 	write_seqlock(&xtime_lock);
 
 	switch (time_state) {
@@ -196,17 +207,17 @@
 	case TIME_INS:
 		timekeeping_leap_insert(-1);
 		time_state = TIME_OOP;
-		printk(KERN_NOTICE
-			"Clock: inserting leap second 23:59:60 UTC\n");
 		hrtimer_add_expires_ns(&leap_timer, NSEC_PER_SEC);
 		res = HRTIMER_RESTART;
+
+		leap = -1;
 		break;
 	case TIME_DEL:
 		timekeeping_leap_insert(1);
 		time_tai--;
 		time_state = TIME_WAIT;
-		printk(KERN_NOTICE
-			"Clock: deleting leap second 23:59:59 UTC\n");
+
+		leap = 1;
 		break;
 	case TIME_OOP:
 		time_tai++;
@@ -219,6 +230,15 @@
 	}
 
 	write_sequnlock(&xtime_lock);
+	
+	if (leap) {
+	    mod_timer(&call_clock_was_set_timer, jiffies);
+	    
+	    printk(KERN_NOTICE
+		    "Clock: %s leap second 23:59:%u UTC\n",
+		    leap == -1 ? "inserting" : "deleting",
+		    leap == -1 ? 60 : 59);
+	}
 
 	return res;
 }
0
rweikusat (2830)
7/17/2012 6:30:55 PM
On Mon, 16 Jul 2012, Jaydeep Chovatia wrote:

> Initially I though this is because of pthread_lock, unlock, condition 
> timedwait, condition wait, etc. system calls I am using in the 
> application, to prove that I have overridden pthread_lock, unlock, 
> condition timedwait, condition wait, unlock, etc. using LD_PRELOAD 
> functionality and concluded that not many threads are waiting here 
> (mutex->__data.__nusers remains less than 5 most of the time), so it 
> doesn?t seem that these sys calls are part of the problem.

I think what you describe might amount to high futex contention. ISTR 
NPTL/futexen are optimized for the uncontended case (= no one to wait for 
when acquiring, no one to wake when releasing), and they are purposely (or 
at least knowingly) heavy-weight in the contended case.

.... The references I can manage now are 
<http://man7.org/linux/man-pages/man7/futex.7.html> and 
<http://kerneltrap.org/node/422> (search the latter for "the main 
application cost is the lost paralellism" for the argument).

I assume __nusers is the number of blocked threads. "Less than 5 most of 
the time" accomodates "more than 3 most of the time", which could mean the 
pthread mutex operations take a trip to the kernel too very frequently.

This could be a good question for comp.programming.threads. Maybe look 
around in <http://www.1024cores.net/home/lock-free-algorithms>.

Laszlo
0
lacos2 (200)
7/17/2012 9:08:55 PM
On Tue, 17 Jul 2012, Rainer Weikusat wrote:

> This affects anything using a subsecond timeout together with 
> CLOCK_REALTIME hrtimer and the reason is these hrtimers maintain a 
> redundant copy of 'the current time' which is not updated from the leap 
> second insertion code because Thomas Gleixner killed the call that was 
> supposed to do this in 2007 (746976a301ac9c9aa10d7d42454f8d6cdad8ff2b).

Thanks for the reference.

<troll>

"Quite a few people consider me to be one of the Grumpy Old Men. That's 
related to my age and the age-related unwillingness to cope with crap."

https://www.linux.com/news/special-feature/linux-developers/591039-30-linux-kernel-developers-in-30-weeks-thomas-gleixner

If he thinks he needs an excuse, he's gonna need a better one.

</troll>

Laszlo
0
lacos2 (200)
7/17/2012 9:16:48 PM
"Ersek, Laszlo" <lacos@caesar.elte.hu> writes:
> On Tue, 17 Jul 2012, Rainer Weikusat wrote:
>> This affects anything using a subsecond timeout together with
>> CLOCK_REALTIME hrtimer and the reason is these hrtimers maintain a
>> redundant copy of 'the current time' which is not updated from the
>> leap second insertion code because Thomas Gleixner killed the call
>> that was supposed to do this in 2007
>> (746976a301ac9c9aa10d7d42454f8d6cdad8ff2b).

[...]

> "Quite a few people consider me to be one of the Grumpy Old
> Men. That's related to my age and the age-related unwillingness to
> cope with crap."
>
> https://www.linux.com/news/special-feature/linux-developers/591039-30-linux-kernel-developers-in-30-weeks-thomas-gleixner
>
> If he thinks he needs an excuse, he's gonna need a better one.

I don't think so. Even the smartest guy will occasionally make silly
mistakes. Because of this, there should be some kind of 'code review
process' by 'other smart guys' to catch these. And the problem is that
this doesn't exist, not the least because critcizing "I'm the core
maintainer of this, that and something else" guys is a really unwise
thing to do for anyone lower in the hierarchy with any ambitions
beyond "My problem is fixed. I'm happy".
0
rweikusat (2830)
7/18/2012 8:18:45 PM
Rainer Weikusat <rweikusat@mssgmbh.com> wrote:
> "Ersek, Laszlo" <lacos@caesar.elte.hu> writes:
> > On Tue, 17 Jul 2012, Rainer Weikusat wrote:
> >> This affects anything using a subsecond timeout together with
> >> CLOCK_REALTIME hrtimer and the reason is these hrtimers maintain a
> >> redundant copy of 'the current time' which is not updated from the
> >> leap second insertion code because Thomas Gleixner killed the call
> >> that was supposed to do this in 2007
> >> (746976a301ac9c9aa10d7d42454f8d6cdad8ff2b).

> [...]

> > "Quite a few people consider me to be one of the Grumpy Old
> > Men. That's related to my age and the age-related unwillingness to
> > cope with crap."
> >
> > https://www.linux.com/news/special-feature/linux-developers/591039-30-linux-kernel-developers-in-30-weeks-thomas-gleixner
> >
> > If he thinks he needs an excuse, he's gonna need a better one.

> I don't think so. Even the smartest guy will occasionally make silly
> mistakes. Because of this, there should be some kind of 'code review
> process' by 'other smart guys' to catch these. And the problem is that
> this doesn't exist, not the least because critcizing "I'm the core
> maintainer of this, that and something else" guys is a really unwise
> thing to do for anyone lower in the hierarchy with any ambitions
> beyond "My problem is fixed. I'm happy".

Code review probably would never have caught this. Catching this bug would
have required a thorough understanding of the timer infrastructure
(especially the niche cases of tick adjustments). I mean, the reviewer would
have had to foresee this kind of error (a missing line in some far flung
module) and know to look for it. Very few other developers would have had
this expertise. It's an unintended consequence of specialization; nobody
around to fix the other guy's mistakes.

IME code review is usually more a waste of time than anything. You usually
need someone as experienced or more experienced than the author wrt to the
subject code. But that means the cost of the review is quite high; higher
than most people are willing to regularly accept. Afterall, there's a reason
the other experienced engineer is off doing something else.

I'm not saying it can't be done systematically. It just requires a
considerable investment in time. Maybe the best way to handle code reviews
is to pair programmers together. Each generally goes about his own business,
but part of his responsibility is to keep track of what the other guy is
doing (keeping his opinions to himself, lest egos get in the way). When he's
finally tasked with reviewing some particuarly hairy bit of code he can come
up to speed immediately. Sort of like a President and Vice-President
situation. The VP keeps his mouth shut, and in return the President keeps
him apprised of everything that's going on so the VP can immediately step
into his shoes when required.
0
William
7/18/2012 9:41:37 PM
William Ahern <william@wilbur.25thandClement.com> writes:
> Rainer Weikusat <rweikusat@mssgmbh.com> wrote:
>> "Ersek, Laszlo" <lacos@caesar.elte.hu> writes:
>> > On Tue, 17 Jul 2012, Rainer Weikusat wrote:
>> >> This affects anything using a subsecond timeout together with
>> >> CLOCK_REALTIME hrtimer and the reason is these hrtimers maintain a
>> >> redundant copy of 'the current time' which is not updated from the
>> >> leap second insertion code because Thomas Gleixner killed the call
>> >> that was supposed to do this in 2007
>> >> (746976a301ac9c9aa10d7d42454f8d6cdad8ff2b).
>
>> [...]
>
>> > "Quite a few people consider me to be one of the Grumpy Old
>> > Men. That's related to my age and the age-related unwillingness to
>> > cope with crap."
>> >
>> > https://www.linux.com/news/special-feature/linux-developers/591039-30-linux-kernel-developers-in-30-weeks-thomas-gleixner
>> >
>> > If he thinks he needs an excuse, he's gonna need a better one.
>
>> I don't think so. Even the smartest guy will occasionally make silly
>> mistakes. Because of this, there should be some kind of 'code review
>> process' by 'other smart guys' to catch these. And the problem is that
>> this doesn't exist, not the least because critcizing "I'm the core
>> maintainer of this, that and something else" guys is a really unwise
>> thing to do for anyone lower in the hierarchy with any ambitions
>> beyond "My problem is fixed. I'm happy".
>
> Code review probably would never have caught this. Catching this bug would
> have required a thorough understanding of the timer infrastructure
> (especially the niche cases of tick adjustments).

Not at all. Employing some common sense had been entirely sufficient:
If clock_was_set is really useless when the clock is being set because
of a leap second, why isn't it also useless when the clock is being
set via settimeofday? 

And these are the kind of questions someone who is not immersed in the
details of a specific problem is likely to ask: Granted, this is a
tree. Coming to think of it, a lot of trees seem to be all around
here. Could we be in a forest?
0
rweikusat (2830)
7/19/2012 4:12:24 PM
Reply:

Similar Artilces:

high sys cpu on cpu# based on mpstat, but if using prstat -Lm -P 7, little cpu is used
The following prstat show little cpu is used on CPU 7# test$> prstat -Lm -P 7 PID USERNAME USR SYS TRP TFL DFL LCK SLP LAT VCX ICX SCL SIG PROCESS/LWPID 4020 oracle 6.0 1.3 0.0 0.0 0.0 2.4 90 0.2 2K 33 18K 0 mysqld/ 9 417 root 0.0 0.0 0.0 0.0 0.0 0.0 100 0.0 12 0 60 6 in.mpathd/1 4020 oracle 0.0 0.0 0.0 0.0 0.0 0.0 100 0.0 6 0 55 0 mysqld/ 1 1022 root 0.0 0.0 0.0 0.0 0.0 0.0 100 0.0 0 0 0 0 lcfd/1 843 root 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 0 fmd/18 879 root 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0 0 0 0 syslogd/52...

High CPU utilization on C3700 ???
My 3700 router keeps getting high CPU utilization (average 50 - 60 %). When it reached 80% or higher, most web sites got time-out and we lost internet connection. Bandwidth monitor from TimeWarner showing usage just about 40% (4 Mbs/10 Mbs): e0: in from Timewarner telecom 10 Mbs e1: out connnected to 4507 with 21 sub interfaces Config has been changed for years. Few access lists run to prevent sub-int talk together. If you know any reason may lead to CPU high, please let me know. I want to try all solutions befor it gets worse. Thank you in advance! Find out which process is eating ...

Accessing CPU utilization in UNIX
Hi all, I'm working on a real-time, database project under UNIX written in C and need to access some value for the CPU utilization. I understand that the kstat library is useful for this purpose, but not exactly how. The kstat chain contains the information I need, but how do I know which name to look for in the chain and how to get the right data? Tips would be greatly appreciated. Thanks in advance. ...

futex high system utilization
Hi, In my c++ (Linux) application I am seeing high CPU utilization (almost 90%)= and it that sys-cpu:user-cpu ratio is 8:2. I then tried to run =93strace= =94 command and found that around 80% of time is being spent in =93futex=94= system call, please see snippet of strace here: % time seconds usecs/call calls errors syscall ------ ----------- ----------- --------- --------- ---------------- 80.66 100.007743 1567 63806 22761 futex 9.15 11.340274 96925 117 4 restart_syscall 5.60 6.939598 4401 1577 pol...

named and high CPU utilization
Hi The named (not very high loaded one) goes taking 100% CPU and can't be stopped in an usual fashion. The same named configuration, on other machine, doesn't behaves like this. It seems that this is only occurring on Windows systems with dual cpu�s. I did find a forum online for a software called Plesk that uses named.exe that has the same issue and the fix that they applied was to assign cpu affinity to named.exe in the code. Windows will allow you to assign cpu affinity through task manager but it won�t allow you to assign affinity to a service! The following is taken from the Plesk forum: "Bind is configured to work on 1 CPU and it makes it do not waste CPU resources." Also I pulled the following from the Oreilly windows 2000 performance guide: "It should not be a big surprise to learn that one secondary effect of multiprocessor coordination and serialization is that it makes caching less effective. This, in turn, reduces the processor's instruction execution rate. To understand why SMPs impact cache effectiveness, we soon take a detour into the realm of cache coherence. From a configuration and tuning perspective, one intended effect of setting up an application to run with processor affinity is to improve cache effectiveness and increase the instruction execution rate. Direct measurements of both instruction execution rate and caching efficiency, fortunately, are available via the Pentium count...

High CPU utilization with Pylab
I am trying to plot a runtime graph using pylab. But the very purpose of functionality is lost because of high CPU percentage hogged by plotting the graph. Here is the piece of code which I have written. def timerfunc(ulcm, dlcm): count=0 xaxis=[] yaxis=[] ion() while 1: xaxis.append(count) dr=(((dlcm.sum)/dlcm.timer)*8) dlcm.counter=0 yaxis.append(dr) dlcm.sum=0 h=plot(xaxis,yaxis,'b') draw() count=count+1 time.sleep(dlcm.timer/1000.0) if (count>100): ...

cpu utilization remain high
Dear All, The cpu utilization of my box remains high....any known reason. CPU states: 6.4% idle, 17.9% user, 33.5% kernel, 42.2% iowait, 0.0% swap thanks and advance. atif76@gmail.com wrote: > Dear All, > The cpu utilization of my box remains high....any known reason. > CPU states: 6.4% idle, 17.9% user, 33.5% kernel, 42.2% iowait, 0.0% swap With such vast information, I can tell you your computer is waiting on some IO event. /Marcin In article <In0Qd.37572$k4.729170@news1.nokia.com>, Marcin Dobrucki <Marcin.Dobrucki@TAKETHISAWAY.nokia.com> wrote: > at...

High CPU utilization by ActiveX32.exe
I have been experiencing very strange behavior on my Win2K system. When I try to run my Norton AntiVirus program or regedit these programs only appear for approx 5-15 seconds and then disappear without any message. The icon for NAV in the systray also appears with an exclamation mark and when my cursor gets near the icon it disappears as well. I receive messages related to my Office Assistant saying it could not be started or to run setup and "Repair Office 2000". Also my computer bogs down and in Task Manager the Activex32.exe process shows as much as 80% CPU. I cannot terminat...

High CPU utilization on Cisco 1760
Hello, sometimes our Cisco 1760 has high CPU utilization and it lasts for 5-20 seconds Here's how it looks rtr-off-02>sh proc cpu hi rtr-off-02 09:56:12 AM Monday Sep 19 2005 msk 55555 3333333333 77777 111 11111111112222244444111113333311111222227777700000 100 ** 90 ** 80 ** 70 ******* 60 ...

ProCurve 2810-48G High CPU utilization
Hi- I'm having an interesting problem with a few new 2810-48G switches. When the 2810 is up-linked to another switch and TAGGED VLANs are allowed on the uplink port, the CPU utilization jumps up to 45 - 55%, and the interfaces get sluggish, even with no devices connected to the 2810 switch. When the uplink connection doesn't have tagged VLANs, just one untagged (normal) vlan, the CPU utilization drops to around 3-5% which is normal I believe. All my switches are ProCurve, and when I replace the 2810 in question with a 2824, it performs perfectly. However, connecting any devices (servers, PCs, etc.) to the 2810 when it has this high CPU utilization causes lots of dropped packets, higher ping times, and general network sluggishness. Has anyone seen anything like this before? thanks, erik PS - Here is some more info about the setup, and how to cause the CPU utilization to go nuts: Two switches, constant configuration, and two possible wiring options: Parent Switch: 5308XL Child Switch: 2810-48G Vlan Configuration: Parent: * Port E4: VLANs 1,2,19, and 3260 Tagged. None untagged. * Port E2: VLAN 1 untagged, none tagged. Child: * Port 1: VLANs 1,2,19, and 3260 tagged. None untagged. * Port 2: VLAN 1 untagged, none tagged. The parent is the backbone switch to which many things are connected, it handles lots of traffic and works great. The child has nothing connected to it presently except it's link with the parent which I change betwe...

Programmers, Programmers, Programmers, ...
As Steve Balmer correctly stated, while making his monkey dance, it is applications and hence programmers that make a platform. The fact though is that if you want to do professional programming, then Linux is the platform for you. I know that this statement will get the heckels up on a lot of trolls in C.O.L.A, but I have a recent experience that proves this. I am currently working for a Windows only house producing a system that receives and transmits around 1000 telegrams per second in each direction on a UDP socket, translates them into a different format and creates a log entry for each ...

Lotus Notes high CPU utilization with IE6
Has anyone run across a problem with Lotus Notes and IE 6 that has the CPU running high? The process that is using the CPU is the nlnotes.exe. This usually occurs when notes is minimized and IE is in use. It happens randomly and usually last for 10 - 15 seconds, even if notes has no agents running. So far we have tried different versions of notes 5.04, 5.08, 5.012. The systems are running windows XP sp1, IE 6 with latest updates. Thanks for any help!! ...

High X CPU util. for video overlay (Xorg7)
After switching to xorg 7.0 I notice a much higher CPU utilisation when watching movies than before (xorg 6.9), although my players (xine, mplayer, vlc) still use video overlay (XV) for displaying. Xorg CPU usage now goes up to 60% when watching a movie compared to less than 5-10% before with Xorg6.9. Was there any change in configuration that might lead to this? What else could be the problem? (NVidia Geforce4MX) NVidia binary drivers are no option. Xorg.0.log shows no warnings or errors. Thanks, Dominique My System: ********** Nvidia Geforce4MX Debian SID (up to dat...

does a Unix tool to display the CPU load of each cpu from a 16 CPU Maschine exist?
Is there a tool to visualize the Processor-load of each processor in a 16 CPU Maschine Sun Solaris 8 Netra-T12? I would like to see if all 16 processors are used by the OS and on which CPU is what running. cpustat does not help me very much. maximilian max wrote: > Is there a tool to visualize the Processor-load of each processor in a 16 > CPU Maschine Sun Solaris 8 Netra-T12? > > I would like to see if all 16 processors are used by the OS and on which CPU > is what running. > > cpustat does not help me very much. > maximilian > > try mpstat Tom "max" <maximilian.gablinger@swisscom.com> writes: >Is there a tool to visualize the Processor-load of each processor in a 16 >CPU Maschine Sun Solaris 8 Netra-T12? >I would like to see if all 16 processors are used by the OS and on which CPU >is what running. >cpustat does not help me very much. >maximilian mpstat Casper -- Expressed in this posting are my opinions. They are in no way related to opinions held by my employer, Sun Microsystems. Statements on Sun products included here are not gospel and may be fiction rather than truth. > Is there a tool to visualize the Processor-load of each processor in a 16 > CPU Maschine Sun Solaris 8 Netra-T12? > > I would like to see if all 16 processors are used by the OS and on which CPU > is what running. Sun Management Center (http://www.sun.com/sunmanagementcenter) will track per-cpu usage as ...

2650 IP Input High CPU Utilization Problem 318827
I have a 2650 deployed as an edge device in a distributed internet provisioning environment. The router has 2 parallel per-packet load-shared T1 connections with internal DSU WICs and only the single fast ethernet interface for the LAN side that is being trunked to a Cat 4006. The config has been in place for over 2 years without a hitch until a recent ISP change that has created a huge problem. Basically the IP Input process incrementally consumes more and more proc time over about 2 hours to the point where I have to reload the router in order for traffic to pass through. Specifics: IOS...

High CPU load but CPU idle
We have HP-UX 11.0 on N class machines. Apache does not run correctly on this machine and HP has no clue why the OS acts this way. CPU 0 has a very high load but is idle. Cpu states: CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS 0 41.95 0.0% 0.0% 0.0% 100.0% 0.0% 0.0% 0.0% 0.0% Anyone have any ideas what's going on? Any time this happens our only option is to reboot. In article <1171384883.637556.310370@v33g2000cwv.googlegroups.com>, tony & jean <vecchiet@gmail.com> wrote: >We have HP-UX 11.0 on N class machines. >Apache do...

convert CPU load to CPU utilization in %
Hi, Is it possible to convert the cpu load average (5min) got from the output of uptime to cpu utilization in %? If so, how to? Thanks in advance. Dolphin wrote: > Hi, > > Is it possible to convert the cpu load average (5min) got from the > output of uptime to cpu utilization in %? If so, how to? Load average is for the system, not the (a) CPU. -- Ian Collins In article <bb06aac4-258c-453f-9648-85cc4df0121a@g1g2000pra.googlegroups.com>, Dolphin <yc282004@yahoo.com.sg> writes: > Hi, > > Is it possible to convert the cpu load average...

Unix Sys Admin + Unix Engineer
Looking for 1 unix sys admin and 1 unix engineer in downtown Atlanta. Must have solid shell scripting experience, and very strong in Perl. Also strong in Apache or SSL or PHP. It is a 3 month contract to hire position with a web hosting company. ...

[ace-bugs] ACE_Dev_Poll_Reactor in ACE-5.7 causes High CPU utilization and severe loss of performance in server app
ACE VERSION: 5.7 HOST MACHINE and OPERATING SYSTEM: CentOS 5.2 and CentOS 4.6 TARGET MACHINE and OPERATING SYSTEM, if different from HOST: COMPILER NAME AND VERSION (AND PATCHLEVEL): gcc (GCC) 4.1.2 20071124 (Red Hat 4.1.2-42) THE $ACE_ROOT/ace/config.h FILE Automatically generated by ./configure THE $ACE_ROOT/include/makeinclude/platform_macros.GNU FILE BUILD METHOD USED: ../configure CONTENTS OF $ACE_ROOT/bin/MakeProjectCreator/config/default.features (used by MPC when you generate your own makefiles): AREA/CLASS/EXAMPLE AFFECTED: ACE_Dev_Poll_Event: A modified version of Dev_Poll_Reactor_Test.cpp DOES THE PROBLEM AFFECT: EXECUTION Affects execution of any application using ACE_Dev_Poll_Event. SYNOPSIS: ACE_Dev_Poll_Reactor in ACE-5.7 causes High CPU utilization and severe loss= of performance in server app DESCRIPTION: When the same instance of ACE_Dev_Poll_Event that is passed to ACE_Reactor = is also used by 2 or more threads to service requests, CPU utilization of t= he process will jump (according to the number of worker threads used). The= same app runs find when ACE_TP_Reactor is used or when the number of worke= r threads is reduced to 1. REPEAT BY: //Search for : REPRODUCE_BUG //=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=...

High run queue, high idle cpu percentage
I have 4-way where vmstat is showing high run queues, but also showing high idle cpu precentage. The rest of the values appear nominal. Any thoughts? vmstat 5 | sed '1,20d' kthr memory page faults cpu ----- ----------- ------------------------ ------------ ----------- r b avm fre re pi po fr sr cy in sy cs us sy id wa 14 0 2232991 1471674 0 0 0 0 0 0 617 10137 554 4 1 94 1 1 0 2236672 1467938 0 0 0 0 0 0 670 3745 774 6 1 92 1 5 0 2234160 1470396 0 0 0 0 0 0 538 2408 440 12 ...

High run queue, high idle cpu percentage
Hey - I have a werid situation: vmstat shows high r (always r>0, sometimes r>10), but id > 90%. The rest of the values are nominal. Any thoughts on this? kthr memory page faults cpu ----- ----------- ------------------------ ------------ ----------- r b avm fre re pi po fr sr cy in sy cs us sy id wa 1 4 2255490 4077584 0 0 0 67 184 0 668 2521 706 11 2 83 4 3 1 2257699 4075363 0 0 0 0 0 0 721 5420 849 58 9 31 1 2 0 2268197 4064791 0 0 0 0 0 0 973 7567 1480 54 5 40 2 2 0 2268...

High cpu....
I have a 3640 showing around 80% cpu load,(even when not at full capacity). - cef is on - the router never had a high load before - I made some changes recently on my network, (and in this router). - Under "show proc cpu" it appears that the "IP Input" is what is using up the cpu. What does the "IP Input" meassure? Note, I'm thinking the problem is one of the following: - mis-matched speed and duplex with another cisco on my internal network,(100-full on this box, but the gateway is a 10-half cisco 2600). - static routes not setup correctly - over...

CPU utilization
Hi all, I'm working on a databsase project under UNIX written in C and need to access some value for the CPU utilization (the time percentage of its time the CPU is not idle). I understand that the kstat library is useful for this purpose, but not exactly how. The kstat chain contains the information I need, but how do I know which name to look for in the chain and how to get the right data? Tips would be greatly appreciated. Thanks in advance. ...

CPU Utilization?
Please help me out how I get CPU utilization from HP-UX Server? I have to incorporate commands into shell script Pal <palhills7@yahoo-dot-com.no-spam.invalid> wrote: > Please help me out how I get CPU utilization from HP-UX Server? > I have to incorporate commands into shell script. If you are using a shell script, perhaps parsing the output of vmstat is in order. What does this shell script do with the CPU utilization data? rick jones -- oxymoron n, commuter in a gas-guzzling luxury SUV with an American flag these opinions are mine, all mine; HP might not want them anyway.....

Web resources about - futex high sys cpu utilization - comp.unix.programmer

Asteroid Retrieval and Utilization - Wikipedia, the free encyclopedia
The Asteroid Retrieval and Utilization (ARU) mission, also known as the Asteroid Initiative , is a potential future space mission proposed by ...

CO2 Utilization in Chemical Processes
CO2 capture has to be followed by proper utilization in useful processes. This group shares knowledge about existing techniques in Chemical Processes, ...

How packet micro-shaping maximizes bandwidth utilization
Enterprises are increasingly virtualizing IT infrastructure by migrating storage, application, and database servers into cloud/hosted datacenters. ...

How packet micro-shaping maximizes bandwidth utilization
Enterprises are increasingly virtualizing IT infrastructure by migrating storage, application, and database servers into cloud/hosted datacenters. ...

Combining utilization of two items - People's Daily Online
EditedandtranslatedbyLiQian,People'sDailyOnlineWorn-outsilkstockingsarenotsuitabletow

Utilization in H1 2014
The Robert Woods Johnson Foundation has published some interesting data on the utilization of hospital services for the first half of 2014. The ...

CREE: Piper Ups Target to $70 on Lighting Prospects, Utilization
Piper Jaffray’s Jagadish Iyer today offers up the take-aways from a management roadshow with executives of light-emitting diode technology maker ...

A Myth of 100% Utilization
Do you remember your very first job? A junior software developer or a quality engineer. An intern maybe. You were expected to do something. All ...

Industrial Production and Capacity Utilization - Business Insider Deutschland
... warm weather contributed to a decline in industrial production in November Wikimedia Industrial production fell 0.6% while capacity utilization ...

Industrial Production increased 0.4% in September, Capacity Utilization increased
From the Fed: Industrial production and Capacity Utilization Industrial production rose 0.4 percent in September after having fallen 1.4 percent ...

Resources last updated: 3/14/2016 5:29:01 PM