[SOLVED] megabytes in free is not kilobytes/1000 but kilobytes/1024

vincix · 08-20-2018, 12:41 PM

on Centos (ubuntu/debian's version has a different syntax when distinguishing between kilo and kibi and so on):

Code:

free -b
              total        used        free      shared  buff/cache   available
Mem:     3153575936  2029826048   309317632    19070976   814432256   886857728

And the number of kilobytes is the number I'm expecting:

Code:

free -k --si
              total        used        free      shared  buff/cache   available
Mem:        3153575     2039660      299229       19070      814686      876961

It's simply divided by 1000.
But when I run:

Code:

free -m --si

I get:

Code:

              total        used        free      shared  buff/cache   available
Mem:           3079        1991         291          18         795         856

Which is actually the result of 3153575/1024, instead of 3153575/1000. And I don't understand why, if I'm using the same power of 1000.

On the other hand, the kibi/mebi notation (so the 1024 power) is consistent:

Code:

free -k
              total        used        free      shared  buff/cache   available
Mem:        3079664     1980132      303200       18624      796332      868120

free -m
              total        used        free      shared  buff/cache   available
Mem:           3007        1945         284          18         777         836

So 3153575936/1024 = 3079664 - that's consistent.
And 3079664/1024 equals 3007, which is also expected.

So what don't I understand exactly? I'm not sure what's going on.

scasey · 08-20-2018, 01:15 PM

Interesting. Although the man page doesn't say so, the --si option appears to only work in conjunction with the -k option
I note that it doesn't even work with no byte option specified, even tho -kilo is the default.

By definition, computer kilobytes == 1024 (2^10) and megabytes is 1000 kilobytes or 1024000, so I suppose it doesn't make sense to apply the 1000 byte divisor to megabytes, as that would misrepresent the size.

onebuck · 08-20-2018, 01:26 PM

Hi,

Maybe you can look at this to help you understand; https://simple.wikipedia.org/wiki/Mebibyte

Hope this helps.
Have fun & enjoy!

vincix · 08-20-2018, 02:41 PM

@scasey
I don't think that's correct. Bear in mind that I'm not talking about the fact that we often call a kilobyte what is actually a kibibytes and on (mega, giga, etc.)
1 kilobyte = 1000 bytes.
1 kibibyte = 1024 bytes. (2^10)

1 megabyte = 1 milion bytes, i.e. 10^6 bytes.
1 mebibyte = 1024^2 = 1048576 bytes

Quote:

Interesting. Although the man page doesn't say so, the --si option appears to only work in conjunction with the -k option

I don't agree with this sentence either, because it would follow that free -m --si and free -m are the same, which isn't true. free -m --si ends up outputting what would seem a values that shouldn't exist. If 1 megabyte is 1 milion bytes, then the correct value in megabytes should be (in my case) 3153575936 divded by 10^6! Not by 1000 and then by 1024! It doesn't make any sense... If you could address this problem directly, I'd greatly appreciate it, instead of pointing me to links that I've read and reread so many times. That's why I'm writing here, so that someone explains exactly what is going on.

Thanks in advance.

[later edit:]
http://extraconversion.com/data-stor...-to-bytes.html
Please look at the screenshot. It follows that the value of free -m --si doesn't make any sense! The value should be 3153, not 3079!

vincix · 08-20-2018, 04:31 PM

Another example in fdisk -l which doesn't seem to be following the same mathematical rules as free does:

Code:

fdisk -l:
Disk /dev/mapper/centos-swap: 2164 MB, 2164260864 bytes, 4227072 sectors

So fdisk says that 1MB equals 10^6 bytes, not 10^3 * 1024 bytes. Which is to be expected. Free says otherwise. This is insanely stupid, in my opinion... Let alone the fact that on Centos they call 'megabytes' what are actually 'mebibytes' and on Ubuntu the same free tool explicitly says mebibytes (-m) and --mega actually refers to the international system, but as you've seen, it doesn't actually, as it multiplies the bytes with 10^3 * 1024...

scasey · 08-20-2018, 05:52 PM

Quote:

Originally Posted by vincix

@scasey
If you could address this problem directly, I'd greatly appreciate it, instead of pointing me to links that I've read and reread so many times. That's why I'm writing here, so that someone explains exactly what is going on.

I didn't point you to any links. I merely observed what free was doing on my desktop, which is a CentOS 7.5 installation.
I stand corrected on the behavior of free -m -si. I guess I got lost running examples.

I don't see it as a "problem" It is what it is. So I'm not going to be much help in "resolving" it.
If you think its a bug, report it. Version and reporting address from my desktop. Reporting address is at the bottom of the man page.

Code:

free -V
free from procps-ng 3.3.10
REPORTING BUGS
       Please send bug reports to ⟨procps@freelists.org⟩

vincix · 08-21-2018, 06:30 AM

Sorry I got so worked up, but, as I didn't really understand what was going on and I myself was struggling to understand it, I found it annoying that I ended up with links to some basic information instead of an actual answer. Yes, I wasn't referring to you in particular, it was a general "you", but the general was that I wasn't being understood. Anyway, thanks for confirming. That's really helpful. At least I know I'm not crazy, given how easy it is (for me) to get tricked by 1024/1000 powers