keeping drives cool in a MicroATX case

[Daniel Lenski]

Take a look at linux/Documentation/laptops/laptop-mode.txt, or google
for something like "linux laptop mode". It's a script that sets the VM
system to reduce paging and remount disks with noatime.

Jerry

Wow... it works! My drives actually spin down now. Cool. But they also
take 3-4 seconds to spin-up, and it's surprisingly audible. I'll have to
tweak this a bit.

Thanks for the tip!

Dan

 

[Tom Stephenson]

Drives aren't all the same but universally it's better to keep them
cool. Most drives have a max temperature of 55C. Doesn't mean they will
fail if there is some short term problem and they go higher than 55 but
running then for an extended period beyond 40 isn't a good idea.

I have seen immediate failures in the temperature
range 70-75C. The drives worked again after cooled down.
I agree that <= 40C in normal operation is a good
rule of thumb.

Generally, more airflow is better. But be careful about the fan(s). If
you run case fans at a full 12V you get maximum performance but minimum
[fan] life.

refer to the fan datasheet for that. there are fans out there
that get >100.000 h lifetime under full speed, e.g. by SmattCooler
and Enermax. They are a bit more expensive though.

I find that fans don't always perform to the datasheet spec (especially
where the MTBF spec is concerned). So I normally want to use a high
quality mainstream fan supplier but even the best of them will have
issues from time to time. Spin it down a bit and the life of the fan is
greatly extended. BTW, fans don't like to be in high temperature
environments either. The MTBF is derated significantly as the
temperature goes up. Same with power supplies.

I have never had trouible with a HDD from fan vibration. Should
mot be an issue today, unless you have a very flimsy case.

I have had tons of trouble with drives and vibration. Consider the
environment of the hot swap drive in a server. The drive is in a
removable mount. It must be secure yet be able to move when the lever
has been activated. And normally there are multiple drives mounted
adjacent to one another. In that environment the spinning mass of a
drive can easily impact the performance of the adjacent drive just as
the fans that are hard mounted in the chassis. The chassis and the drive
carrier have to be very carefully designed to avoid issues with drives
being impacted by vibration.

Now you may have had trouble with drives being influenced by vibration
and not know it. Drives make errors when they read the data from the
media. There are sufficient ECC bits encoded into the record to correct
the data as it is read from the media. But when the data is so badly
scrambled the data cannot be corrected by the ECC logic the drive has to
do a re-read. That slows performance cause you have to take another
revolution to fetch the data. Will the data come out OK the second time
it is read? Maybe, maybe not. What about the third time? How many times
will the drive read the data before it gives up and declares an error is
a feature of that individual drive design (firmware on the drive). But
the drive may take many many retry events before it declares a sector to
be probationary. (pending sector).

When I evaluate the performance of a drive I read the entire surface of
the drive block by block and measure the time it takes to retrieve the
block. I have my own code that I use for this purpose but there are
commercial software packages that do the same thing.

Many performance issues can be identified as vibration issues at the source.

Scan the data from the drive while mounted in the chassis. Scan the data
again from the same drive while mounted externally on a very heavy mass
not connected to the chassis. Compare the read performance. The delta in
the numbers will the the contribution of the chassis (vibration mostly
but temperature as well).

A good chassis design and a good hard drive will have near identical
performance. But I have seen combinations where the performance is off
by 20 - 30 percent. Those products never were released to production.

Hitachi also has a low power line and normal Samsungs asre also
pretty close to low power. In addition, if performance is less
of an issue, 2.5" notebook drives are an option. Most are at <2W
under load, whioch puts them at about 20-30% power of even the
low-power 3.5" drives.

Arno

My experience is with enterprise class drives from WD, Hitachi and
Seagate. All three are very good at what they do. I understand that the
new Sammy drives are very good as well.

Things are going to get tough in the hard drive environment because of
the SSD products coming our way. As the capacity and price features
improve, the SSD will begin to impact the hard drive business.

Tom S.

 

[Squeeze]

Tom Stephenson wrote in news:[email protected]

Drives aren't all the same but universally it's better to keep them
cool. Most drives have a max temperature of 55C. Doesn't mean they will
fail if there is some short term problem and they go higher than 55 but
running then for an extended period beyond 40 isn't a good idea.

I have seen immediate failures in the temperature
range 70-75C. The drives worked again after cooled down.
I agree that <= 40C in normal operation is a good
rule of thumb.

Generally, more airflow is better. But be careful about the fan(s). If
you run case fans at a full 12V you get maximum performance but minimum
[fan] life.

refer to the fan datasheet for that. there are fans out there
that get >100.000 h lifetime under full speed, e.g. by SmattCooler
and Enermax. They are a bit more expensive though.

I find that fans don't always perform to the datasheet spec (especially
where the MTBF spec is concerned). So I normally want to use a high
quality mainstream fan supplier but even the best of them will have
issues from time to time. Spin it down a bit and the life of the fan is
greatly extended. BTW, fans don't like to be in high temperature
environments either. The MTBF is derated significantly as the
temperature goes up. Same with power supplies.

I have never had trouible with a HDD from fan vibration. Should
mot be an issue today, unless you have a very flimsy case.

I have had tons of trouble with drives and vibration. Consider the
environment of the hot swap drive in a server. The drive is in a
removable mount. It must be secure yet be able to move when the lever
has been activated. And normally there are multiple drives mounted
adjacent to one another. In that environment the spinning mass of a
drive can easily impact the performance of the adjacent drive just as
the fans that are hard mounted in the chassis. The chassis and the drive
carrier have to be very carefully designed to avoid issues with drives
being impacted by vibration.

Now you may have had trouble with drives being influenced by vibration
and not know it. Drives make errors when they read the data from the
media. There are sufficient ECC bits encoded into the record to correct
the data as it is read from the media. But when the data is so badly
scrambled the data cannot be corrected by the ECC logic the drive has to
do a re-read. That slows performance cause you have to take another
revolution to fetch the data. Will the data come out OK the second time
it is read? Maybe, maybe not. What about the third time?

How many times will the drive read the data before it gives up and de-
clares an error is a feature of that individual drive design (firmware on
the drive).

And then the driver takes over making it do that all over again, several times.

 

[Arno Wagner]

Drives aren't all the same but universally it's better to keep them
cool. Most drives have a max temperature of 55C. Doesn't mean they will
fail if there is some short term problem and they go higher than 55 but
running then for an extended period beyond 40 isn't a good idea.

I have seen immediate failures in the temperature
range 70-75C. The drives worked again after cooled down.
I agree that <= 40C in normal operation is a good
rule of thumb.

Generally, more airflow is better. But be careful about the fan(s). If
you run case fans at a full 12V you get maximum performance but minimum
[fan] life.

refer to the fan datasheet for that. there are fans out there
that get >100.000 h lifetime under full speed, e.g. by SmattCooler
and Enermax. They are a bit more expensive though.

I find that fans don't always perform to the datasheet spec (especially
where the MTBF spec is concerned).

MTBF has no relation to lifetime, or rather the lifetime
is where the MTBF gives the failure probability. So MTBF is
not at all for wear and tear but for "random" failures.

So I normally want to use a high
quality mainstream fan supplier but even the best of them will have
issues from time to time. Spin it down a bit and the life of the fan is
greatly extended. BTW, fans don't like to be in high temperature
environments either. The MTBF is derated significantly as the
temperature goes up. Same with power supplies.

The enermax ones have 100'000h lifetime at 80C. Smartcooler have
100'000h at 40C, if I remember correctly. Things like NMB
not high quality.

I have had tons of trouble with drives and vibration. Consider the
environment of the hot swap drive in a server. The drive is in a
removable mount. It must be secure yet be able to move when the lever
has been activated. And normally there are multiple drives mounted
adjacent to one another. In that environment the spinning mass of a
drive can easily impact the performance of the adjacent drive just as
the fans that are hard mounted in the chassis. The chassis and the drive
carrier have to be very carefully designed to avoid issues with drives
being impacted by vibration.

I have had up to 12 Disks in one server tower. Never had an issue.

Now you may have had trouble with drives being influenced by vibration
and not know it. Drives make errors when they read the data from the
media. There are sufficient ECC bits encoded into the record to correct
the data as it is read from the media. But when the data is so badly
scrambled the data cannot be corrected by the ECC logic the drive has to
do a re-read. That slows performance cause you have to take another
revolution to fetch the data. Will the data come out OK the second time
it is read? Maybe, maybe not. What about the third time? How many times
will the drive read the data before it gives up and declares an error is
a feature of that individual drive design (firmware on the drive). But
the drive may take many many retry events before it declares a sector to
be probationary. (pending sector).

Sorry, no. Not with said server or the 22 attached cluster nodes
(whith 2-3 disk each in one disk cage) during 3 years.

When I evaluate the performance of a drive I read the entire surface of
the drive block by block and measure the time it takes to retrieve the
block. I have my own code that I use for this purpose but there are
commercial software packages that do the same thing.

I have used the same approach before the smartmontools for linux
began to work well. No observations with regard to vibration.

Many performance issues can be identified as vibration issues at the source.

Scan the data from the drive while mounted in the chassis. Scan the data
again from the same drive while mounted externally on a very heavy mass
not connected to the chassis. Compare the read performance. The delta in
the numbers will the the contribution of the chassis (vibration mostly
but temperature as well).

A good chassis design and a good hard drive will have near identical
performance. But I have seen combinations where the performance is off
by 20 - 30 percent. Those products never were released to production.

Aha. Maybe I just bought too good quality to observe the effect.
But then, many of the disks were Maxtors.

My experience is with enterprise class drives from WD, Hitachi and
Seagate. All three are very good at what they do. I understand that the
new Sammy drives are very good as well.

Things are going to get tough in the hard drive environment because of
the SSD products coming our way. As the capacity and price features
improve, the SSD will begin to impact the hard drive business.

Not for many apps. SSDs are far from competitive, if seek time
and throughput is secondary. And they will remain more expensive
than HDDs for quite some time. Might even be for decades.

Arno