Hacker News Reader

Comment by signa11 | Hacker News Reader

just curious: would these machines have non-rotational media only ?

I imagine that the hard drives themselves wouldn't be filled with the oil - perhaps wrapped in a heat-conducting plastic beforehand, or maybe drives are stored outside of oil altogether.

Does anyone have any statistics on what contributes the most heat? I'm assuming power supply and processor/GPU are the worst offenders.

Edit: yup, the drives are wrapped:

> First, our technicians remove server fans and put them aside for future use or reverse-modification at a later date. Then, each hard drive is encapsulated using GRC’s proprietary hard drive sealing system. Finally, the CPU thermal grease is replaced with a non-soluble foil.

It's in the FAQ.

"hard drive sealing system" - I'm pretty sure you can't completely seal a disk, it needs some surrounding air to deal with pressure changes and such.

Wonder if they can control the pressure to the point you can seal it... or if the seal itself incorporates an air flow system.

How complex can this sealing system be while still allowing the disks to fit in the bays?

I'm all but certain disks are sealed. If air has a path to the platters then so does dust.

Air molecules are not dust-sized.

You can use a bladder to quickly equalize air pressure without exchanging air. (Air will slowly penetrate most bladders, but it's not fast enough for rapid pressure changes.)

Some heaters use a similar approach to limit corrosion. http://www.westank.com/bladder-tanks.php

That's what I was thinking too, and the pressure changes shouldn't be so drastic that the bladder would burst.

Pure speculation: The hole is for, as you said, pressure differences. The head uses air pressure to float above the platter. Changes in pressure could bring the head closer/farther away from the platter which could have unwanted consequences. Pressure fluctuations are due in part to changes in temperature. Since the temperature is regulated by the oil, and that temperature will change more slowly than with air, plugging the hole is relatively safe.

SSDs in server farms are the way to go anyway.

the price-points for the two are way off the scale e.g. on newegg, a 120gb ssd is expensive than 1tb 7200rpm sata..

But you get tens of thousands IOPS instead of hundreds

Which may be important, or may be utterly worthless. It's like "total cost of ownership" FUD, numbers are useless outside their original context, and everybody has a different context.

This. For many high perf applications, anything outside of RAM (yes, even with SSD's) is a no-no.

Wow.. really? Is this a trend with any documentation? Seems to me that there are still a lot of mentions of SSD failures floating around the tubes, and that plus the significant cost for large scale storage over traditional platters would make it a very tough sell to any organization wanting to deploy large quantities of servers.

Mechanical disk supply chains are increasing platter drive costs significantly while at the same time improvements in reliability and cost for SSDs are bringing costs lower. And yes, there have been large deployments: http://www.computerworld.com/s/article/9218811/EBay_attacks_...

Reply on news.ycombinator.com