Hardware monitoring, Dell 2650, Debian

po0k

I've been looking into this for the last few days in work and I'm stumped.
I can't seem to find any references to what's in the servers.
Talking to one of the guys who tried it before, he said that there's an Adaptec monitoring chip onboard, but for which there are no source drivers, only a precompiled RH binary (for RH7.x afaik). He said it was pretty much useless.
Does anyone know the chip model/family which are used in these servers? We have some bigger dell boxes too, 1U-4U, but I don't know what's in them.
Googling and trawling through Dell's site reveals nothing.

My aim is to poll the monitoring chips every 5minutes or so for system/cpu temps, voltages and fan rpm speeds (for the meantime) through SNMP and use MRTG to graph the output.

I've gotten some basic disk health monitoring functionality from using smartmontools (deb package name) which gives me some SMART info and drive temps, and while that's nice and all, I'd like to know when the system is overheating.

Any ideas?

nadir

does i2c + acpi , lmsensors not work ?

po0k

As far as I know it's a fairly stripped down kernel that's running on the machines, so peripheral support is minimal.
How would ACPI help with basic monitoring?
I installed lm_sensors, but during the setup process it was looking for modules which weren't installed.
I would go messing around with the kernel but it's *kinda* in production, I've got to check with those responsible if I can tinker around a bit more.

I've no idea what HW mon chip/chips are used in it, except that the 3rd party chip is made by Adaptec. I forget what circuitry is in the northbridge and/or southbridge.

nadir

have a look in /proc/acpi/ if its installed.
i have

Grrr]nadir:~/docs>ls /proc/acpi/
ac_adapter battery dsdt event fan power_resource sleep
alarm button embedded_controller fadt info processor thermal_zone

Im afraid though if you cant rebuild the kernel with those extra modules for lmsensors, you are stuck. But if acpi is running properly you ought to be able to throw some simple scripts together to gather info.
My only knowledge of adaptec is with scsi cards, dont know of any specific monitoring hardware.

flamegrill

John,

I'd be highly supprised if anything on the mainboard of that dell is adaptec over the SCSI controller(s).

Typically its ServerWorks or Plain old Intel chipsets. we have a few dells and thats whats in em

Paul

po0k

I'm going by what I've been told by one of the lads here who's previously tried to reverse engineer it.
I whipped off the cover a while ago and came accross the ServerWorks chipset (CSB5 s'bridge afaik, not 100% sure of n'bridge, PCI ID didn't match with anything I could see on plasma-online.de.
Afaik the CSB5/6 are based on the Intel PXII3 or 4, and afaik these have some degree of HW monitoring don't they? The pxii4 driver in 2.6.0+ now has support for both the CSB5 and 6 so I *may* be out of this soon enough
I'm going to fiddle around and see what I can pull from it.

Then all that remains is for me to get a better grasp on writing stuff for snmp

po0k

right, I've upgraded to 2.6.7, added in support for *everything* to do with i2C bar the debug messages and I've added support for the southbridge
(ServerWorks CSB-5) and the Intel PIIX3/4

I'm trying to track down information regarding the HW-mon capabilities of the piixn (which the CSB-5 is a clone of). Nothing much so far.

This is an excerpt from dmesg output:
i2c /dev entries driver
i2c-parport: using default base 0x378
piix4-smbus 0000:00:0f.0: Found 0000:00:0f.0 device
i2c-sis96x version 1.0.0


I've installed lm-sensors, libsensors1 and sensorsd, plugging away at it.
If anyone has any ideas or if they can tell me if I'm on a wild goose-chase I'd appreciate it

Gerry

Ok, what I've found is that the pe 2650 normally comes with a DRAC III card, which provides a vnc server for remote console access, but also provides hardware monitoring. It can be accessed by installing dell openmanage software, and hence through the os, but it also has its own nic. If you connect it up, then there should be a webserver up and running. ( I think the default ip is 192.168.0.20 ).

Remotely, you can use perl to query this web server and get results back in a nice format. Have a look here: http://www.lanceerplaats.nl/PowerEdge/RAC/

I've not had a chance to test this out yet, hopefully will do soon. Let us know how you get on.

Gerry

Ok, I tested this out, works fine. I just hooked up a crossover, and browsed to
https://192.168.0.120
192.168.0.120 is the default address, it will be displayed when the machine is booting up, but I'd say yours will be default as well.
Theres a lot of java and such crap used to make the monitors refresh, so it would be adviseable to use the perl module.
I also was able to get a remote console, pretty slow but it worked

po0k

Thannks for that phil, I had wondered about those but assumed they were just emergency recovery console ports.
I piddled around with it this morning, but didn't get past the login screen as the guy who set them up wasn't around today.

I finally managed to get 2.6.7 running on the 2560, but the modules for i2c support never compiled, so I'm going to try them statically as a proof-of-concept.


Has anyone had any experience with lm-sensors under 2.6?

po0k

Right, got 2.6.7 running, modules compiled.

When I run sensors-detect (from lm-sensors), it picks up a PII4X compatible sensor (Serverworks CSB5 soutbridge in this case), loads i2c-pii4x, then loads i2c-dev.
Everything's looking fine right?

Then I get this:

Next adapter: SMBus PIIX4 adapter at 08a0 (Algorithm unavailable)
Do you want to scan it? (YES/no/selectively): yes

Machine hangs

I saw lm-sensors mutter something about making sure "/dev/sysfs is mounted" or something to that effect.
The CSB5 is supported by the i2c-pii4x module, as it's just a clone of the "real" PII4X.
The annoying thing is I'm not even sure what sensors, if any, are in the CSB5.

Is there any way of using that DRACIII card to ist system health information under debian linux?

Gerry

Well, if you look at the second url in my post, you can query it remotely with perl. Or if you want to get the info without an extra patch lead, you can install dell's openmanage software: http://linux.dell.com/monitoring.shtml

po0k

Cool, didn't see that.

Do you need a windows front-end manager for it or can I snmp request data off it?

/me reads the manual before asking anymore questions