Excessive fan noise on Sun Fire x4200

[David Kirkby]

I've got a Sun Fire x4200 (2U high), and have installed Solaris 11.
But the thing is incredibly noisy. Far more so than a 1 U IBM x3550
server, despite the IBM has a lot more processing power.

Unlike many of you here, I've not spent my working life in server
rooms, but there's a complaint from someone else about the noise of
this

http://forums.techsoup.org/cs/community/f/23/p/29941/105798.aspx

where the person says he has configured loads of HP, Dell and IBM
boxes, but not found anything as loud as this x4200.

I've used:

$ ipmitool sensor
(not sure if I needed to be root to run that or not. The machine is
off now)

and find the fans are running at about 5000-6000 rpm in a room about
21 deg C in my house. They are making a huge amount of noise, but the
components inside are barely warm. The heat sinks on the CPUs are
barely above blood temperature.

As a test, I removed 4 of the 6 fans at the front. The noise level
reduced a fair amount, and cooling was still adequate. The speed of
two working fans did not increase, suggesting to me that whatever
temperature sensor was adjusting the fan speed was happy with just two
fans running. If two fans running at 6000 rpm can keep the machine
cool, why the hell does it need 6 of them going at that speed?

It's obvious the fan speed control is working, as running a very CPU
intensive benchmark, or blocking a lot of the air intake, does result
in the fan speed increasing.

I've got a friend staying here with me, who has worked a lot in server
rooms, and he says although servers are noisy, the x4200 is
excessive.

I think if I kept this machine (which is looking unlikely), i'd
probably stick some resistors in series with the fans to slow them
down a bit.

Does anyone have any less drastic ways to reduce the fan speed?

Obviously 6 fans provides redundancy, so me removing 4 of the 6 is not
a long term solution, but as a test it proved to me that the machine
is excessively cooled.

Dave

[ChrisQ]

On 01/15/12 09:29, David Kirkby wrote:
>
> Obviously 6 fans provides redundancy, so me removing 4 of the 6 is not
> a long term solution, but as a test it proved to me that the machine
> is excessively cooled.
>
> Dave

One of the reasons why sun h/w is so reliable is that it's properly sorted
in terms of temperature testing. Iirc. they are usually rated to 35c
ambient,
but will have been development tested to a much higher level. I remember at
dec css in Reading being involved in the testing of a piece of kit that had
dozens of thermocouples fitted inside every orifice of the box,
including psu.
It was wheeled into a test chamber and cycled well beyond the specified
range
for weeks, running diagnostics and with temperatures data logged. The only
thing that failed was the disk in the floppy drive. Such testing is the only
way to ensure reliability and is one of the reasons why generic kit can be
made and sold much cheaper: By skimping on hardware quality (especially
psu)
and validation testing. That's if there's any at all in some vendors
product.

If the box is designed to run to 35c and you are sure that the machine will
always be at much less that that, then yes, you can probably get away
with fewer
or lower speed fans. Do remember though that nothing kills electronics
faster
that wide range temperature cycling and excessive temperature, so make
sure the
remaining fans are circulating air to the right areas. Another point is
that
the firmware or Solaris may have settings for fan speeds ?.

Still think an ml350 or ml370 (quiet fans) would have been a better bet,
but
ymmv :-). All this stuff can be a journey of discovery, but that's part
of the
fun of it as well...

Regards,

Chris

[David Kirkby]

On Jan 15, 1:17 pm, ChrisQ <m...@devnull.com> wrote:
> On 01/15/12 09:29, David Kirkby wrote:
>
>
>
> > Obviously 6 fans provides redundancy, so me removing 4 of the 6 is not
> > a long term solution, but as a test it proved to me that the machine
> > is excessively cooled.
>
> > Dave
>
> One of the reasons why sun h/w is so reliable is that it's properly sorted
> in terms of temperature testing. Iirc. they are usually rated to 35c
> ambient,

But is is more reliable than decent HP or IBM kit?

I know the Sun SPARC 20 did not suffer from over-cooling. Quite the
opposite!

> but will have been development tested to a much higher level.

Yes, I appreciate that. I wish all companies would do that. I know
many where the items don't even meet the specs they claim.

> If the box is designed to run to 35c and you are sure that the machine will
> always be at much less that that, then yes, you can probably get away
> with fewer
> or lower speed fans.

But if the box is designed to run up to 35 deg C, I can understand why
it has a lot of fans. But there's no need to run them so fast so much
of the time. I don't see what that can do for reliability - it just
wears the fans out quicker.

> Do remember though that nothing kills electronics
> faster
> that wide range temperature cycling and excessive temperature, so make
> sure the
> remaining fans are circulating air to the right areas.

I don't really want to run with fewer fans. But the noise from this
machine is in a league of its own.

> Another point is
> that
> the firmware or Solaris may have settings for fan speeds ?.

That I'm not sure. Since its Sun hardware running Solaris, I would
have assumed that its not an issue with a lack of a driver. Certainly
the speeds do change as I increase the work the machine does.

> Still think an ml350 or ml370 (quiet fans) would have been a better bet,
> but
> ymmv :-).

I did look at that, but decided against it.

I've bought a couple of IBM servers, which are more what I wanted for
Linux.

* IBM x3550 (2 x quad core Xeon 3.16 GHz, 1U)
* IBM x3650 (1 x quad core Xeon 2.XX GHz, 2U)

The first IBM is very quick indeed - the second one will be when I've
transplanted some of the bits from the 1U unit to the 2U unit. For
reasons I don't know, IBM servers seem a lot cheaper than Dell or HP
on eBay. There's no way you can get 8 cores in 3 GHz Xeon processors
for sub £400 on eBay if you buy Dell or HP. But you can if you go with
IBM, the hardware seems to attract less interest.

My plan is to move the two 3.16 Xeons in the 1U x3550 to the 2U
x3650, as that will take 48 GB RAM. With 8 cores that will give me
something decent. I've got 48 GB RAM on order, along with the
necessary heatsink and VRM.

Both the IBMs are too noisy to have in a house, but my intention was
to put the IBM(s) + Sun in the garage. (Luckily my garage is detached
and some distance from the house!) But the Sun just seems excessively
noisy. Performance wise it is far inferior to the IBMs, but make one
hell of a lot more noise.

> All this stuff can be a journey of discovery, but that's part
> of the
> fun of it as well...

True. And not as frustrating as trying to install commercial software
that uses FlexLM.

I always promised myself I'd never touch a bit of software that needs
FlexLM, but I might be forced to. But I'm having a struggle installing
it. The software is supposed to be supported on Redhat 4 and 5, but
the license manager appears to be supported only on Redhat 3 and 4!!!
So I've just downgraded the IBM from CentOs (i.e Redhat) 5.6 to CentOS
4.7 so I can install both the software and license manager on the same
machine.

> Regards,
>
> Chris

Dave

[Nomen Nescio]

> One of the reasons why sun h/w is so reliable is that it's properly sorted
> in terms of temperature testing. Iirc. they are usually rated to 35c
> ambient, but will have been development tested to a much higher level.

Intel is telling people to run their data centers at 100F ambient whatever
that is in European ;-) I'll put a Sun box up against anything Intel makes
any day of the year. Excessive fan noise is goodness. If you can't stand the
noise don't run a server!

[ChrisQ]

On 01/15/12 14:55, David Kirkby wrote:

>
> But is is more reliable than decent HP or IBM kit?
>
> I know the Sun SPARC 20 did not suffer from over-cooling. Quite the
> opposite!
>

Have never run ibm servers, but they are an old school h/w vendor and they
will have all the processes in place for product development. Would
think that they are very thorough. Dec -> Compaq -> hp used to be the same,
but one wonders how the consumer electronics mentality has diluted the hp
brand. Thankfully, they spun off the test equipment division, but have heard
reports that even that isn't built to the same standard anymore. My lab is
full of old hp kit: Mid 80's onwards, even the odd 70's item and kit from
that period just seems to work forever. World class product is about
attention
to detail, which is expensive if done right.

> I always promised myself I'd never touch a bit of software that needs
> FlexLM, but I might be forced to. But I'm having a struggle installing
> it. The software is supposed to be supported on Redhat 4 and 5, but
> the license manager appears to be supported only on Redhat 3 and 4!!!
> So I've just downgraded the IBM from CentOs (i.e Redhat) 5.6 to CentOS
> 4.7 so I can install both the software and license manager on the same
> machine.
>

Have worked on sites where tool vendors used flexlm. It's fine when
it works, but if the license manager is on a server and the server goes
down, all development with the tools stops, which can cost K's for a large
development group. It was quite common with embedded
tool vendors in the early 90's and one site I worked at had real problems
with the vendor when the server crashed and lost a load of files.
The vendor flatly refused to issue temporary licenses until we got the
server
back up and running. This client was a blue chip co as well, not
some small outfit, so you would expect to be able to pull some rank, but
no luck. The other point being, of course, that any vendor using flexlm is
essentially saying that everyone is dishonest and that they don't trust
you, so
why should I trust their ethics either ?. Feel the same way about h/w
dongles and have
always insisted that we negotiate a non dongled license or look elsewhere.
Most vendors want the business and are flexible if the sale depends on
it and /
or twist their arm a bit :-). Of course, the rise of open source tools for
embedded development has eliminated the need for commercial tools in
many cases...

Regards,

Chris

[David Combs]

So if you were the tool vendor, and whose product was totally unique
and saved you a whole bunch of money or enabled things you could
do no other way, and you (vendor) wanted to avoid angry customers
when that single server went down -- what approach would you take?

(Assume that yes people would steal the software in a flash, if they
had a chance.)

Would you put flexlm on TWO servers? More?

And dongles -- (a) what main disadvantages go with them?


(b) How difficult for someone to clone one?


-----

What is it that Microsoft does to validate software on your computer?

And, how well does it work (for them)?


Thanks,

David

[Doug McIntyre]

dkc...@panix.com (David Combs) writes:
>(Assume that yes people would steal the software in a flash, if they
>had a chance.)

Generally, what I've seen of software piracy is that the general
random user that pirates some software package most likely would never
have bought it in the first place. The biggest prevention of piracy is
to make sure a company using the software is using the correct # of
copies of the software that they have bought.

But then this breaks when IT has to reload the workstation. Some
companies reimage workstations constantly..

>Would you put flexlm on TWO servers? More?

If it was that critical.

>And dongles -- (a) what main disadvantages go with them?

Dongles get lost. They get broken. There are incompatibilities with them.
Back in the days of parallel ports, they could block the printer from working.
More than one could block the other from working. In the USB days, if
you need to license 10 software pieces all with dongles, how are you
going to plug them all in? Are large USB hubs going to be standard
issue for you?

Dongles need drivers. Are you going to walk your end user through
installing random company's 64-bit drivers that are needed for the new
workstation that came out after you shipped your software?

If you have a $10k piece of software, and you have a lost dongle, how
likely is the company going to believe you or not (ie. see my 1st point)?

>(b) How difficult for someone to clone one?

It is very difficult to clone one. Much easier to crack the software.
Ie. code sequences like this are very common..

jmp.l DongleAPI(MAGIC1,MAGIC2,MAGIC3)
cmp 0x432139593
bne exit(1)

Hmm, lets change that to 'beq' and see what happens? MAGIC strings are
well known.

>What is it that Microsoft does to validate software on your computer?

That depends on what software you are talking about, and what
generation. Latest windows OS uses a crypto hash to generate the
random letters that are needed for install, and then the software
also needs to activate back to Microsoft over internet or phone to
make sure that it is an install key that was actually issued and not
used before.

>And, how well does it work (for them)?

How well? Like almost all software is released by the pirates before
it is shipped to the stores, cracked and ready to go?
Granted, I think Office 2010 actually took the pirates maybe a few months
to have a rock-solid bypass method..