On May 28, 2019, at 21:27 , bzs@theworld.com <bzs@TheWorld.com> wrote:
On May 28, 2019 at 19:56 owen@delong.com <mailto:owen@delong.com> (Owen DeLong) wrote:
It’s unlikely to apply to much of anything in a datacenter other than disks.
Ok, disks, a mere bagatelle of a concern.
Then again obviously disks have gotten much, much better about thermal change since people in, e.g., temperate climates might take their laptops' running disks from a long, frigid walk into a warm building.
There was a time when you weren't supposed to move a disk while it was still spinning (e.g., hot swaps had to be able to cut power before removal so you could give the heads several seconds to stop and park), not sure how they solved that so completely, again, laptops.
A big part of solving this was smaller platters. Remember, platters from the time you’re talking about were somewhere between 5.25” diameter and 14” or even 19” diameter in some cases. Platters were usually made of fairly rigid steel and the heads were separated from the platters by air cushions and Bernouli’s principal and very very little else. Today’s disk drives have the heads much closer to the media, but the media is a lot less sensitive to brief head contact. The media is often on thin flexible plastic substrates and the separation of the heads is often purely mechanical. Head insertion and parking mechanisms have changed quite a bit as well, allowing for much sturdier gantries (the head assembly is now usually mounted on a triangular gantry with an arced side where the actuators connect. This allows a significantly larger amount of material to be used in constructing the gantry and the heads travel in an arc across the media instead of in a linear motion as was common in older drives.
The reason it applies to disks is because rapid cooling of a drive will lead to uneven cooling of the platters which may cause abnormal stresses leading to shattering and/or warpage (depending on the material the drive platters are made from).
Also head clearances and other moving parts tolerances.
Well, the primary thing that’s going to cause you grief for abrupt temperature changes is when the platters warp, resulting in reduced head clearance (possibly even negative head clearance). In the case of some glass platters, shattering will also make for a really bad day.
Most electronic components can tolerate a pretty steep thermal curve in either direction so long as the curve doesn’t take them out of spec one way or the other.
Also, most circuit boards and the like do not have enough mass to surface area ratio to lead to significant temperature differentials within a small physical distance.
Then again if you're cooling a room from, say, 115F to 70F you only need one excuse to consider the rate of cooling and disks would be a pretty good excuse.
SSDs no doubt are obsoleting even that concern.
But I still tend to worry about the relationship of resistance to temperature in circuits as a general principle tho perhaps in the likely range it's not a major concern.
Modern resistors don’t tend to move as much as in the past. However, even for that, as long as you don’t exceed the operating temperature thresholds at either end, you should be fine. The rate of cooling/heating isn’t really an issue for that.
Anyhow, IT'S WORTH A THOUGHT if something extreme happens to temperatures in your machine room.
If you get out of tolerance, then there are lots of things to think about. If you’re within allowed operating range for your equipment, lots less. Rate of change, as I said, is the one that’s unique to disk drives due to the high mass to surface area ratio and the tight mechanical tolerances.
You might not want to fling open the doors and windows of a 110+F room to 0F outside air and begin turning everything back on as the room's air thermometer begins to register 70F a few minutes later.
Yep… If you’ve got spinning media, that’s probably not the best plan. For solid state stuff, it’s probably OK as long as you don’t leave the door open until the room starts to drop below 10ºC.
Water condensation can also be a concern, after a prolonged A/C failure it may be hot and humid in the room depending on the climate etc.
Truth. If you create a new weather pattern inside the datacenter, you’re conducting some form of experiment likely to yield “interesting” results. Owen
File Under: MORE THINGS TO WORRY ABOUT!
Owen
On May 28, 2019, at 12:18 , bzs@theworld.com wrote:
Something to keep in mind is that some equipment, disks in particular, should only be cooled at a certain rate once they're hot, often annoyingly slow by the specs like 2-3 degrees C per hour but there are probably circuits sensitive to this also which could be anywhere.
It came up because it happened to me in Cambridge, MA in the dead of winter and every helpful person in the building came by to suggest I just open windows and doors to the snowy outdoors to get things running sooner.
It should be in the specs and if you're concerned about equipment running in too hot an environment you might be concerned about this also. Particularly after a forced power-down which also powers down equipment fans while the chips etc are still hot so will continue heating cases.
Ambient air temperature might not be telling you the whole story is the point.
I keep one of those big 5' fans, looks like something they use in Hollywood for windstorms and feels a bit like it on high, for just this sort of reason tho even if I just think it's getting warm, and several smaller fans to point at racks etc.
The best thing you can do if it gets too hot is keep the air moving.
(Where to plug the fans in after a power shutdown is your problem, I knew someone would think that!)
-- -Barry Shein
Software Tool & Die | bzs@TheWorld.com | http://www.TheWorld.com Purveyors to the Trade | Voice: +1 617-STD-WRLD | 800-THE-WRLD The World: Since 1989 | A Public Information Utility | *oo*
-- -Barry Shein
Software Tool & Die | bzs@TheWorld.com <mailto:bzs@TheWorld.com> | http://www.TheWorld.com <http://www.theworld.com/> Purveyors to the Trade | Voice: +1 617-STD-WRLD | 800-THE-WRLD The World: Since 1989 | A Public Information Utility | *oo*