[lm-sensors] Power Supply Monitoring

[lm-sensors] Power Supply Monitoring

[Jacob Steinberger]

Gents,

Before I get too far into lmsensors, I'm hoping I can get a quick  
yea/nay on the path I'm going.

I'm trying to do power supply monitoring to detect if we're having a  
problem (loss of grid power, loss of UPS power, voltage issues, etc).  
I've found that some wiki pages state that this is, to some degree,  
the point of lmsensors. What I'm failing at figuring out if I can do  
this on my hardware or not.

I've run sensors-detect, and I get back ...

Driver `coretemp':
   * Chip `Intel Core family thermal sensor' (confidence: 9)

Driver `ipmisensors':
   * ISA bus, address 0xca8
     Chip `IPMI BMC KCS' (confidence: 8)

Driver `i5k_amb':
   * Chip `Intel AMB FB-DIMM thermal sensor' (confidence: 9)

Does either ipmisensors or i5k_amb support voltage type statistics? If  
not, I would assume if my hardware does have these sensors, lmsensors  
doesn't support them.

Thanks in advance for any responses (or lack there of ;),

Jacob


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Mark E. Hansen]

On 06/02/09 14:49, Jacob Steinberger wrote:
> Gents,
> 
> Before I get too far into lmsensors, I'm hoping I can get a quick  
> yea/nay on the path I'm going.
> 
> I'm trying to do power supply monitoring to detect if we're having a  
> problem (loss of grid power, loss of UPS power, voltage issues, etc).  
> I've found that some wiki pages state that this is, to some degree,  
> the point of lmsensors. What I'm failing at figuring out if I can do  
> this on my hardware or not.

I meant to copy my response to the lm-sensors list and missed it
the first time. Sorry for the duplicate.

With regard to the UPS monitoring, you would use something like
NUT for that: <http://www.networkupstools.org/>

With that, you can ask for all kinds of information about the
state of the UPS, including line voltage, etc. (depending on
the features of the UPS of course).

Note that I use NUT on Linux. I don't know if it runs on Windows,
but as you didn't mention your OS, I'm just taking a guess  :-)

I can't answer about whether your specific sensor chips are
supported by libsensors, so I'll leave that for someone else
to answer.

> 
> I've run sensors-detect, and I get back ...
> 
> Driver `coretemp':
>    * Chip `Intel Core family thermal sensor' (confidence: 9)
> 
> Driver `ipmisensors':
>    * ISA bus, address 0xca8
>      Chip `IPMI BMC KCS' (confidence: 8)
> 
> Driver `i5k_amb':
>    * Chip `Intel AMB FB-DIMM thermal sensor' (confidence: 9)
> 
> Does either ipmisensors or i5k_amb support voltage type statistics? If  
> not, I would assume if my hardware does have these sensors, lmsensors  
> doesn't support them.
> 
> Thanks in advance for any responses (or lack there of ;),
> 
> Jacob
> 
> 
> _______________________________________________
> lm-sensors mailing list
> lm-sensors@lm-sensors.org
> http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
> 



_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Matt Roberds]

On Tue, 2 Jun 2009, Jacob Steinberger wrote:
> I'm trying to do power supply monitoring to detect if we're having a
> problem (loss of grid power, loss of UPS power, voltage issues, etc).

The voltage readings you get from lm-sensors are usually of the DC
output of the computer power supply.  The readings also tend to have a
relatively low update rate - maybe a few times a second max.  If your
input AC power is really bad, you might be able to see it by looking at
the output, but if the AC power only glitches out occasionally, you
probably won't see it this way.

UPSes tend to have various amounts of AC power monitoring available,
and that might be a better place to look.  Network UPS Tools
http://www.networkupstools.org/ runs on Linux and can talk to many
types of UPSes.

Beyond that, the next step is usually to go to the site and look for
any obvious problems - do half the machines reboot when the machine
shop downstairs fires up the arc welder?  Did somebody set up 42U of
servers on a 100 foot 16 gauge (30 m 1.0 mm^2) extension cord?  Then
you start poking at it... depending on how subtle the problem is and
your budget, you start with a $30 Kill-a-Watt meter, then move up to
a $200 Fluke multimeter, and then maybe a multi-thousand-dollar Fluke
or Dranetz data logger.

> Does either ipmisensors or i5k_amb support voltage type statistics?

As far as I can tell, i5k_amb is just a temperature sensor.  If you have
any voltage sensors (which is likely), they will come through the
ipmisensors driver.  If you've ever been into the BIOS setup screens of
this machine, and one of them displayed the power supply voltages, there
is a decent chance that ipmisensors will be able to show you those
voltages as well.

Standard disclaimers apply; I don't get money or other consideration
from any companies mentioned.

Matt Roberds


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Jacob Steinberger]

My hope is that if one power supply is no longer working, thus no longer 
producing any DC output, I could alarm on that value being 0. Again, 
this is just a hope.

> UPSes tend to have various amounts of AC power monitoring available,
> and that might be a better place to look.  Network UPS Tools
> http://www.networkupstools.org/ runs on Linux and can talk to many
> types of UPSes.
>
Tend to, but since the guy that does the monitoring rarely has the 
ability to make the purchases of said device, you tend to end up with 
old refurbished devices that don't have the ability to be monitored 
without a laptop and a console cable.

> Beyond that, the next step is usually to go to the site and look for
> any obvious problems - do half the machines reboot when the machine
> shop downstairs fires up the arc welder?  Did somebody set up 42U of
> servers on a 100 foot 16 gauge (30 m 1.0 mm^2) extension cord?  Then
> you start poking at it... depending on how subtle the problem is and
> your budget, you start with a $30 Kill-a-Watt meter, then move up to
> a $200 Fluke multimeter, and then maybe a multi-thousand-dollar Fluke
> or Dranetz data logger.
>
I'm really just looking for a way to do a simple function check on the 
power supplies to detect when they're operational. Not necessarily to 
find out what voltage is going through, but to see if it's even working 
or not. For example, we have one power supply on grid, the other is on 
UPS - the easiest way to detect if there's a problem with UPS or grid 
power is to monitor the power supplies on the boxes. This would function 
as a dual monitor if there was ever an issue with just a power supply.

Jacob

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Jean Delvare]

On Tue, 2 Jun 2009 22:00:00 -0500 (CDT), Matt Roberds wrote:
> On Tue, 2 Jun 2009, Jacob Steinberger wrote:
> > I'm trying to do power supply monitoring to detect if we're having a
> > problem (loss of grid power, loss of UPS power, voltage issues, etc).
> 
> The voltage readings you get from lm-sensors are usually of the DC
> output of the computer power supply.  The readings also tend to have a
> relatively low update rate - maybe a few times a second max.  If your
> input AC power is really bad, you might be able to see it by looking at
> the output, but if the AC power only glitches out occasionally, you
> probably won't see it this way.

Actually this is only partly true. While the sysfs interface
rate-limits the readings, the chip itself is usually sampling the
voltage at a much higher rate, and alarms which are raised stick until
they are read. So you may not be able to see the voltage value at
glitch time, but you should definitely see an alarm for it the next
time you check it.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Mark E. Hansen]

On 06/02/09 20:09, Jacob Steinberger wrote:
> My hope is that if one power supply is no longer working, thus no longer 
> producing any DC output, I could alarm on that value being 0. Again, 
> this is just a hope.
> 
>> UPSes tend to have various amounts of AC power monitoring available,
>> and that might be a better place to look.  Network UPS Tools
>> http://www.networkupstools.org/ runs on Linux and can talk to many
>> types of UPSes.
>>
> Tend to, but since the guy that does the monitoring rarely has the 
> ability to make the purchases of said device, you tend to end up with 
> old refurbished devices that don't have the ability to be monitored 
> without a laptop and a console cable.

You seem convinced that NUT won't work. I'm not sure why you've
come to this conclusion. Perhaps you've researched NUT lately
and found that it doesn't support any of the UPSs you're using.
Of course, that would have to be pretty bad luck, as NUT supports
many (most?) of them.

> 
>> Beyond that, the next step is usually to go to the site and look for
>> any obvious problems - do half the machines reboot when the machine
>> shop downstairs fires up the arc welder?  Did somebody set up 42U of
>> servers on a 100 foot 16 gauge (30 m 1.0 mm^2) extension cord?  Then
>> you start poking at it... depending on how subtle the problem is and
>> your budget, you start with a $30 Kill-a-Watt meter, then move up to
>> a $200 Fluke multimeter, and then maybe a multi-thousand-dollar Fluke
>> or Dranetz data logger.
>>
> I'm really just looking for a way to do a simple function check on the 
> power supplies to detect when they're operational. Not necessarily to 
> find out what voltage is going through, but to see if it's even working 
> or not. For example, we have one power supply on grid, the other is on 
> UPS - the easiest way to detect if there's a problem with UPS or grid 
> power is to monitor the power supplies on the boxes. This would function 
> as a dual monitor if there was ever an issue with just a power supply.

How do you expect software on a computer to tell you when the power
supply on that computer has gone to zero volts?


> 
> Jacob
> 
> _______________________________________________
> lm-sensors mailing list
> lm-sensors@lm-sensors.org
> http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
> 


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Jacob Steinberger]

> You seem convinced that NUT won't work. I'm not sure why you've
> come to this conclusion. Perhaps you've researched NUT lately
> and found that it doesn't support any of the UPSs you're using.
> Of course, that would have to be pretty bad luck, as NUT supports
> many (most?) of them.

The goal isn't to monitor the UPS, it's to monitor the power supplies  
on the individual servers. Just monitoring the UPS isn't sufficient if  
you have a power supply failure.

> How do you expect software on a computer to tell you when the power
> supply on that computer has gone to zero volts?

Because it should know the voltage coming of each power rail? These  
servers have dual power supplies, and I'm looking to detect a failure  
with *one*. Detecting a failure in both is easy - the box is offline.

Jacob


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Power Supply Monitoring

[Jean Delvare]

On Wed, 03 Jun 2009 07:31:21 -0700, Mark E. Hansen wrote:
> On 06/02/09 20:09, Jacob Steinberger wrote:
> > I'm really just looking for a way to do a simple function check on the 
> > power supplies to detect when they're operational. Not necessarily to 
> > find out what voltage is going through, but to see if it's even working 
> > or not. For example, we have one power supply on grid, the other is on 
> > UPS - the easiest way to detect if there's a problem with UPS or grid 
> > power is to monitor the power supplies on the boxes. This would function 
> > as a dual monitor if there was ever an issue with just a power supply.
> 
> How do you expect software on a computer to tell you when the power
> supply on that computer has gone to zero volts?

Some computers have dual (or more) power supply units. On these
machines, as long as at least one PSU is still up and running, you can
monitor them all. However it is relatively rare that mainboards are
wired to monitor that kind of setup.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors