News Liquid Cooled HDD Study Touts Greater Reliability, Lower TCO

[Admin]

Iceotope worked with Meta to convert air-cooled, HDD-packed servers to precision single-phase immersion cooling - showing wide-ranging benefits including lower TCO and greater reliability.

Liquid Cooled HDD Study Touts Greater Reliability, Lower TCO : Read more

 

[InvalidError]

Immersion cooling for HDDs sounds a bit overkill when drives only dissipate 10-15W each with most of that heat sunk into the cast aluminum frame. You could probably achieve most of the benefits by simply using the drives' mounting holes and some thermal paste to attach coolant channels to the side walls which could do double-duty as drive rails.

 

[thestryker]

I'd imagine pretty much any improving the quality of air cooling would end up decreasing density which is problematic for datacenter. The biggest question I have regarding this is simplicity of swapping a failed drive, or lack thereof.

The fact that they ended up seeing better reliability by keeping more consistent temps is interesting and something I'm going to keep in mind when I put together my replacement server.

 

[tmwrnj]

Immersion cooling for HDDs sounds a bit overkill when drives only dissipate 10-15W each with most of that heat sunk into the cast aluminum frame. You could probably achieve most of the benefits by simply using the drives' mounting holes and some thermal paste to attach coolant channels to the side walls which could do double-duty as drive rails.

Unfortunately, that approach doesn't scale.

If you've got 60 drives per 4u chassis, that modest 10-15W per drive works out at 7,200W to 10,800W per rack. You can try to conduct that heat into the rack hardware, but it isn't going to go anywhere without a bunch of forced air. When you do blow that heat out of the rack then it has to go somewhere, which in a data center means the HVAC system. You're then using a whole bunch of energy to exhaust that air from the building or chill it and recirculate it.

Immersion cooling for HDDs sounds like overkill, but it's a potentially rational approach when you're putting tens of thousands of HDDs in the same room. The efficiency and reliability benefits might be outweighed by capital and maintenance costs, but it's a question worth serious consideration for customers like Meta.

 

[InvalidError]

You can try to conduct that heat into the rack hardware, but it isn't going to go anywhere without a bunch of forced air.

What part of integrating liquid cooling into drive rails requires forced air?

 

[richardvday]

What part of integrating liquid cooling into drive rails requires forced air?

You're still reducing density is I believe what was meant. It most definitely would reduce the density depending on how thick they needed to be.

 

[richardvday]

Surprise ! Reducing heat leads to longer product life who would have thought 😂

 

[InvalidError]

You're still reducing density is I believe what was meant. It most definitely would reduce the density depending on how thick they needed to be.

it is ~15W tops, cooling channels would need to be about as small as they can possibly be made without compromising structural integrity or becoming a clogging problem. If you go with immersion, you still need space between drives for coolant flow along with something to hold them in place, so I'd imagine the density would be quite similar.

Surprise ! Reducing heat leads to longer product life who would have thought 😂

Except that according to BlackBlaze, their HDD failure rates go up when temperature is under 25C, so HDDs don't like being too cold either.