PowerEdge 1900 fan noise solution

You have a server that is designed to keep on cooling effectively with every drive slot filled with a 15krpm drive, every card slot filled with a high current draw card, and CPUs and memory being worked to capacity - with enough cooling spare for nothing to be run under any thermal stress.

 

That said, I've no direct experience with the 1900. Looking at the Hardware Owners' Manual I see that the airflow is not 'straight through' for many of the fans, also six fans have been used.

 

 

I've got a 2950 III in my development rig that has more in it than your 1900 (two quad core processors versus your one - though admittedly my processors are 45nm E5430, six SAS drives - two 15krpm and four 7200rpm, and eight DIMMs versus your two - there's also a PERC 6/i with its battery, SAS 5/E card and a DRAC 5 in there). That lot has four main fans - which are redundant enough to cope with the failure of one - plus a fan in each of the power supplies. The airflow is 'straight through' the case, as is typical for a rack server - air in at the front and out at the back.

 

I wouldn't be surprised if my 2950 III is quieter than your 1900 - and, as you say, you expect rack mount servers to be loud. That said, the R200 that I use as a firewall is pretty quiet.

 

 

I understand your complaint - but your options are limited. The fans Dell have chosen probably aren't the quietest and the BIOS is likely optimised for good cooling rather than low noise. That said, I would check to see that you have the latest BIOS on your machine.

 

Changing the fans will void your warranty, as will running the server with one or more fans removed.

 

 

I suspect Dell have studied the airflow on this system fairly carefully - you don't hear of cooling problems with servers. I would contend that you simply don't have full evidence to say that the fans you propose have adequate airflow - though in your relatively lightly loaded system you'll probably get away with it.

 

 

The connector is some sort of header - though it could quite possibly be a custom one made for Dell. If you are determined to change the fans, I'd order up a new set of fans from Dell spares, and disassemble them for the connectors, wiring and holders. You can then solder the wires to the new fan. The part number is shown on the label that I can't quite read in your picture - under the top bar code you have, I believe, CN_0C9887_... - anyway, the part number is the C and the next four digits.

 

There will be a ground and 12V power on that connector - the others two pins might be a PWM and a tacho line if you're lucky - but you'll still have to find fans that have similar specifications for those two lines. An oscilloscope will help greatly in sorting out what's what if the data sheet of the Dell fan doesn't help.

 

 

This could be hard work, fairly expensive, and I wonder whether Dell will entertain any warranty claims caused by cooling issues.

 

 

I wonder about an alternative approach - contact Dell, explain your issue and that the 1900 may not be the correct choice for you. I doubt you will get any success, but you could be surprised.

I know that Dell has a reason to put all those fans in and how limited my options are regarding the warranty, but it still doesn't explain why Dell hasn't thought it would be reasonable to pwm-control the fans according to the actual temperature of the system. It feels funny they haven't thought about the fact that it IS after all advertised as a "small office" server which can't always be tucked away in a separate sound insulated room.

 

Anyway. I bought a set of Nexus PWM fans (DF1209SL-3PWM) which have a max rpm of 2500 and managed to find out the Dell proprietary pwm pinout:

 

firstly, the Dell connector is propably custom made (no electronic shops I contacted had seen such anywhere nor is it found in any Molex - or other standard catalogs), so i had to solder and cut the wires to replace the fans.

 

The wire output in Nexus Real Silent PWM fan 92mm is the Intel standard:

 

Yellow +12VDC

Black 0 Ground

Green Tachometer out RPM

Blue PWM signal in

Dell's proprietary pinout is:

Red +12VDC

Black 0 Ground

Yellow Tachometer out RPM

Blue PWM signal in

So, after you connect the proper wires and put a PWM fan in every slot the server will boot up normally, recognize and control the fan RPM's and work as expected. A fun fact i noticed was that during cool nights the room temperature drops enough for the fans to slow down to almost 800RPM which triggers the "low rpm alert" on the mobo and sends the fans to full blast for 5 minutes, this happens in cycles if the room temperature is too low (around 20C).

I'm now trying to find out if the PWM signal is completely controlled by the motherboard and not trusted with the OS.

As I said in my first reply, you should be able to get the fans from Dell Parts. The part number is shown in your picture - but isn't quite readable there. On the label with the bar codes, it begins CN-0, then there's a C and four digits. The C and four digits is the part number; call the Dell Parts number for your country and order as many as you need. Indeed, you may find them cheaper online from non-Dell sources.

 

A quick bit of Googling suggests that the part number is C9857 - if I'm right, Googling for Dell C9857 will probably dig up several suppliers.

 

 

Dell probably don't send an on site engineer for a hot swap part; my experience is that they try to avoid sending engineers if they can just send a part. Across various Dell systems over the years I've had a hard disk, a graphics card (well, the fan - but Dell swap the entire card for a remanufactured one, which probably was returned in the first place with a failed fan) and a monitor fail. In each case, once Dell support had ascertained that my fault finding was correct and that I was happy to fit the parts myself, they sent a replacement part with instructions for the delivery driver to collect the faulty part in exchange.

 

They do send an on site engineer when something big needs doing. I've had two motherboard swaps - a workstation motherboard was swapped because a serial port had blown, and a laptop board was swapped because an insufficiently strong Bluetooth module connector needed changing (the Bluetooth module was also replaced). In both cases I got an engineer visit - in fact, it was the same engineer who did both jobs.

 

 

Parts for service calls can be remanufactured - the hard disk was new, as was the monitor and the workstation motherboard (which, judging by the ancient BIOS revision, had been sat in the parts warehouse since that particular model started to ship several years earlier). The graphics card had been repaired, as had the laptop motherboard (the engineer commented that they were changing loads of boards for Bluetooth problems on that particular model; it was just a case of someone fitting the upgraded connector and testing the board).

 

I can't think that they'd bother to rework hot swap fans - the time taken to extract a fan from its housing and fit a new one is likely to cost far more than treating the housing and connector as disposable.

 

 

I expect that the fans are controlled entirely by the BIOS and/or the BMC firmware - if you can influence the behaviour directly from Linux, it will likely be by trying to communicate directly with the fan controller(s) on the motherboard, which will leave you in a permanent fight against the motherboard's fan control.

 

The fan control is likely at a very low level; I doubt, for example, that it can talk to the hard disks to ascertain their temperature. Put a 15krpm SAS drive in each drive bay and that's a lot of heat needing to be carried away. You have six SATA drives, which I would expect to be relatively cool running 7200rpm units - but the motherboard has no way of knowing that. The same is certainly true for expansion cards and memory - the system has no way of knowing those temperatures (though some servers do have an expansion card bay sensor).

 

I doubt that there's more than one or two temperature sensors on the motherboard, plus a sensor on the die of each processor. That isn't a lot of information for the system to decide on a cooling scenario, hence the need to use a worst case scenario.

With the OMSA-repository enabled it's possible to get a few readings from the system with ipmitools and omreport, except that the features listed in the OMSA manual didn't apply to PE 1900. When I typed as root into the console:

omconfig chassis fancontrol speed=quiet

the system responded "No Fan control feature found on this system.". Also trying to set the min and max warning tresholds for the fans proved impossible: "Minimun warning fan treshold setting unavailable on this probe".

 

Seems like these are artificial restrictions in the firmware or software, and I'm going to contact Dell in order to find out what can be done about it.

 

I hope this thread is useful for all the other PE 1900 users I've seen struggling with the noise in various forums and blogs.

 

Also decided to study the situation with IPMI:

ipmitool sensor thresh "FAN 6 RPM" lnr 300

returns "Error setting threshold: Command illegal for specified sensor or record type"

I'd love to get a confirmation from other PE1900 users if they are able to extract any more information from their system than just fan speeds and these sensor readings with ipmitools:

 

        Temp             | 01h | ok  |  3.1 | -51 degrees C
        Temp             | 02h | ok  |  3.2 | 50 degrees C
        Ambient Temp     | 08h | ok  |  7.1 | 29 degrees C
        CPU Temp Interf  | 76h | ns  |  7.1 | Disabled