Hard drive issues

[kubicle]

While on the topic of the actual options, there should be no practical benefit from using 'noatime' option over 'relatime' which I think is the current default.

[SteveRiley]

While on the topic of the actual options, there should be no practical benefit from using 'noatime' option over 'relatime' which I think is the current default.

Linus hates (well, in 2007 anyway) relatime the most, apparently: https://lkml.org/lkml/2007/8/10/200

Of course, he also had somthing else to say: https://lkml.org/lkml/2007/8/4/98

But yeah, "noatime,data=writeback" will quite likely be *quite* noticeable (with different effects for different loads), but almost nobody actually runs that way.

Which, ah, is exactly what I do! LOL

[kubicle]

Those are rather old posts (2007, like you mentioned, relatime was rather "new" back then, and the "critical" post isn't actually from Linus)...

The "relatime" thing that David mentioned might well be very useful, but
it's probably even less used than "noatime" is. And sadly, I don't really
see that changing (unless we were to actually change the defaults inside
the kernel).

Which is what they did (I doubt it would have made a kernel default if linus "hated it").

http://unix.stackexchange.com/questi...de-the-default

EDIT:

But yeah, "noatime,data=writeback" will quite likely be *quite* noticeable

This is compared to atime (strictatime), not compared to relatime. you get basically the same performance boost with relatime (theoretically, noatime is marginally faster, but also comes with a few problems...some programs won't function with noatime)

[FrankBarmentlo]

I would not recommend noop for spinning media. Since noop is just simple FIFO queue, it's likely going to result in much more head movement, thus raising the overall temperature of the drive. CFQ groups I/O requests into batches that minimize seek time, reducing head movement.

well, reducing temperatures is what I wanted to reach, so i guess it's better not to implement this?

and there's a clear difference on windows 8 + VMWare(3 virtual machines), and Linux.. linux' temperatures rise fast, and skyhigh, while windows gets warm, not hand-burning-hot(laptop),

could this be the way the hard drive gets handled by the file system?

[dibl]

(theoretically, noatime is marginally faster, but also comes with a few problems...some programs won't function with noatime)

I vaguely recall there were mutt issues on servers way back when, but is there any current reporting on user-space problems? My google-fu can't find anything recent about it. I've been using the noatime option for years, and haven't seen a problem (that I'm aware of).

Thanks!

[kubicle]

I vaguely recall there were mutt issues on servers way back when, but is there any current reporting on user-space problems? My google-fu can't find anything recent about it. I've been using the noatime option for years, and haven't seen a problem (that I'm aware of).

Problems with noatime are rare, especially on desktop machines (mutt isn't server only, but I'd guess not in terribly wide use). But there is a lot of software floating around, and it's hard to tell which rely on atime and tracking down a possible problem might be quite difficult.

The point I was trying to make is that you should have a good reason to change defaults, and you're unlikely to see any performance improvements with noatime compared to relatime (both will give pretty much the same benefit over strictatime). noatime is not going to reduce disk writes compared to relatime so "These will stop time stamping every time a file is accessed" is somewhat outdated information.

There is nothing wrong in using noatime, if you are aware of possible problems, but you're not going to see the performance improvements you got 5 years back (when atime was default).

[SteveRiley]

and there's a clear difference on windows 8 + VMWare(3 virtual machines), and Linux.. linux' temperatures rise fast, and skyhigh, while windows gets warm, not hand-burning-hot(laptop)

Nearly all power management is handled through ACPI. Hardware manufacturers tune their stuff for Windows, and tweak ACPI accordingly. Many of these tweaks are vendor-specific and poorly documented, if at all. Linux doesn't know about all the various ACPI wonkiness and can't be as effective. Thus, hardware in general runs a bit warmer when you put Linux on it.

[dibl]

...you're not going to see the performance improvements you got 5 years back (when atime was default).

Yep, that would be true enough. Funny (or not) how the valuable things you used to know aren't so valuable as time goes by.

[Teunis]

Don't tell your wife of 20 years

[dibl]

Don't tell your wife of 20 years

Heh -- the BAD news is, my wife of 32 years already figured it out -- can't sneak sneak anything by that one!

The GOOD news (I guess) is, the passage of time affects her about the same.