The entire reason "storage vendors prefer" 1000-based kilobytes is so that they could misrepresent and over-market their storage capacities, getting that 24-bytes per-kb of expectation-vs-reality profit.
It's the same reason—for pure marketing purposes—that screens are measured diagonally.
I'm suprised they didn't mention kibibyte. (Edit: they did) There are plenty of applications where power-of-2 alignment are useful or necessary. Not addressing that and just chastising everyone for using units wrong isn't particularly helpful. I guess we can just all switch to kibibytes, except the HDD manufacturers.
This ambiguity is documented at least back to 1984, by IBM, the pre-eminent computer company of the time.
In 1972 IBM started selling the IBM 3333 magnetic disk drive. This product catalog [0] from 1979 shows them marketing the corresponding disks as "100 million bytes" or "200 million bytes" (3336 mdl 1 and 3336 mdl 11, respectively). By 1984, those same disks were marketed in the "IBM Input/Output Device Summary"[1] (which was intended for a customer audience) as "100MB" and "200MB"
Edit: The below is wrong. Older experience has corrected me - there has always been ambiguity (perhaps bifurcated between CPU/OS and storage domains). "And that with such great confidence!", indeed.
-------
The article presents wishful thinking. The wish is for "kilobyte" to have one meaning. For the majority of its existence, it had only one meaning - 1024 bytes. Now it has an ambiguous meaning. People wish for an unambiguous term for 1000 bits, however that word does not exist. People also might wish that others use kibibyte any time they reference 1024 bytes, but that is also wishful thinking.
The author's wishful thinking is falsely presented as fact.
I think kilobyte was the wrong word to ever use for 1024 bytes, and I'd love to go back in time to tell computer scientists that they needed to invent a new prefix to mean "1,024" / "2^10" of something, which kilo- never meant before kilobit / kilobyte were invented. Kibi- is fine, the phonetics sound slightly silly to native English speakers, but the 'bi' indicates binary and I think that's reasonable.
I'm just not going to fool myself with wishful thinking. If, in arrogance or self-righteousness, one simply assumes that every time they see "kilobyte" it means 1,000 bytes - then they will make many, many failures. We will always have to take care to verify whether "kilobyte" means 1,000 or 1,024 bytes before implementing something which relies on that for correctness.
And a megabyte is depending on the context precisely 1000x1000=1,000,000 or 1024x1024=1,048,576 bytes*, except when you're talking about the classic 3.5 inch floppy disks, where "1.44 MB" stands for 1440x1024 bytes, or about 1.47 true MB or 1.41 MiB.
* Yeah, I read the article. Regardless of the IEC's noble attempt, in all my years of working with people and computers I've never heard anyone actually pronounce MiB (or write it out in full) as "mebibyte".
Interestingly, HD floppies actually are 2 "MB" unformatted without the various overhead. This is how 1.68 "MB" DMF is possible. Extra-high Density (ED) 2.88 "MB" is similarly 4 "MB" unformatted.
It's too late. Powers-of-two won. I'm the sort of person who uses "whom" in English, but even I acknowledge that using "KB" to mean 1,000, not 1,024, can only breed confusion. The purpose of language is to communicate. I'm all for pedantry when it's compatible with clarity, but we can't reconcile the two goals here.
Why don’t kilobyte continue to mean 1024 and introduce kilodebyte to mean 1000. Byte, to me implies a binary number system, and if you want to introduce a new nomenclature to reduce confusion, give the new one a new name and let the older of more prevalent one in its domain keep the old one…
I usually just say kilobyte when speaking, and say “binary kilobyte” or “decimal kilobyte” if it’s not clear from context. I still (usually, but I forget) use the IEC symbols when I mean binary and the SI symbols when I mean decimal. The extra ‘i’ doesn’t cost that much.
I like how the GNU coreutils seem to have done. They use real, 1024-byte kilobytes by default, but print only the abbreviation of the prefix so it's just 10K or 200M and people can pretend it stands for some other silly word if they want.
You can use `--si` for fake, 1000-byte kilobytes - trying it it seems weird that these are reported with a lowercase 'k' but 'M' and so on remain uppercase.
Whenever this discussion comes up I liked to point out that even in the computer industry, prefixes like kilo/mega/etc more often mean a power of 10 than a power of 2:
Agreed. For the naysayers out there, consider these problems:
* You have 1 "MB" of RAM on a 1 MHz system bus which can transfer 1 byte per clock cycle. How many seconds does it take to read the entire memory?
* You have 128 "GB" of RAM and you have an empty 128 GB SSD. Can you successfully hibernate the computer system by storing all of RAM on the SSD?
* My camera shoots 6000×4000 pixels = exactly 24 megapixels. If you assume RGB24 color (3 bytes per pixel), how many MB of RAM or disk space does it take to store one raw bitmap image matrix without headers?
The SI definitions are correct: kilo- always means a thousand, mega- always means a million, et cetera. The computer industry abused these definitions because 1000 is close to 1024, creating endless confusion. It is a idiotic act of self-harm when one "megahertz" of clock speed is not the same mega- as one "megabyte" of RAM. IEC 60027 prefixes are correct: there is no ambiguity when kibi- (Ki) is defined as 1024, and it can coexist beside kilo- meaning 1000.
The whole point of the metric system is to create universal units whose meanings don't change depending on context. Having kilo- be overloaded (like method overloading) to mean 1000 and 1024 violates this principle.
If you want to wade in the bad old world of context-dependent units, look no further than traditional measures. International mile or nautical mile? Pound avoirdupois or Troy pound? Pound-force or pound-mass? US gallon or UK gallon? US shoe size for children, women, or men? Short ton or long ton? Did you know that just a few centuries ago, every town had a different definition of a foot and pound, making trade needlessly complicated and inviting open scams and frauds?
Thank you. I had to scroll way down to find anyone defending using SI prefixes to mean what they mean everywhere else. A decade ago, I decided to alias "du" to "du --si" and not look back. Entire countries have switched from imperial to metric units. Switching to using base 10 for RAM is really just fine.
For all the people commenting as if the meaning of "kilo" was open to discussion... you are all from the United States of America, and you call your country "America", right?
<joke> How to tell a software engineer from a real one? A real engineer thinks that 1 kilobyte is 1000 bytes while software engineer believes that there are 1024 meters in a kilometer :-) </joke>
I had a computer architecture prof (a reasonably accomplished one, too) who thought that all CS units should be binary, e.g. Gigabit Ethernet should be 931Mbit/s, not 1000MBit/s.
I disagreed strongly - I think X-per-second should be decimal, to correspond to Hertz. But for quantity, binary seems better. (modern CS papers tend to use MiB, GiB etc. as abbreviations for the binary units)
Fun fact - for a long time consumer SSDs had roughly 7.37% over-provisioning, because that's what you get when you put X GB (binary) of raw flash into a box, and advertise it as X GB (decimal) of usable storage. (probably a bit less, as a few blocks of the X binary GB of flash would probably be DOA) With TLC, QLC, and SLC-mode caching in modern drives the numbers aren't as simple anymore, though.
It makes it inconvenient to do things like estimate how long it will take to transfer a 10GiB file. Both because of the the difference between G and Gi, and because one is in bytes and the other is in bits.
There are probably cases where corresponding to Hz, is useful, but for most users I think 119MiB/s is more useful than 1Gbit/s.
NAND flash has overprovisioning even on a per-die basis, eg. Micron's 256Gbit first-generation 3D NAND had 548 blocks per plane instead of 512, and the pages were 16384+2208 bytes. That left space both for defects and ECC while still being able to provide at least the nominal capacity (in power of two units) with good yield, but meant the true number of memory cells was more than 20% higher than implied by the nominal capacity.
The decimal-vs-binary discrepancy is used more as slack space to cope with the inconvenience of having to erase whole 16MB blocks at a time while allowing the host to send write commands as small as 512 bytes. Given the limited number of program/erase cycles that any flash memory cell can withstand, and the enormous performance penalty that would result from doing 16MB read-modify-write cycles for any smaller host writes, you need way more spare area than just a small multiple of the erase block size. A small portion of the spare area is also necessary to store the logical to physical address mappings, typically on the order of 1GB per 1TB when tracking allocations at 4kB granularity.
> Explorer is just following existing practice. Everybody (to within experimental error) refers to 1024 bytes as a kilobyte, not a kibibyte. If Explorer were to switch to the term kibibyte, it would merely be showing users information in a form they cannot understand, and for what purpose? So you can feel superior because you know what that term means and other people don’t.
The author doesn’t actually answer their question, unless I missed something?
They go on to make a few more observations, and say finally only that the current different definitions are sometimes confusing, to non experts.
I don’t see much of an argument here for changing anything. Some non experts experience minor confusion about two things that are different, did I miss something bigger in this?
Metric prefixing should only be used with the unit bit. There is no confusion there. I mean, if you would equate a bit with a certain voltage threshold, you could even argue about fractional bits.
Approximating metric prefixing with kibi, Mibi, Gibi... is confusing because it doesn't make sense semantically. There is nothing base-10-ish about it.
The author decidedly has expert syndrome -- they deny both the history and rational behind memory units nomenclature. Memory measurements evolved utilizing binary organizational patterns used in computing architectures. While a proud French pedant might agree with the decimal normalization of memory units discussed, it aligns more closely to the metric system, and it may have benefits for laypeople, it fails to account for how memory is partitioned in historic and modern computing.
I do agree up to a point as I still need to double take when I see MiB, but that said, I also do agree keeping SI unit prefixes standardized has great advantages.
So the "sane" options would be either not using SI for digital, or, what was chosen, change the colloquial prefixes in the digital world. The former would have been easier in the short term.
82 comments
[ 0.19 ms ] story [ 75.9 ms ] threadIt's the same reason—for pure marketing purposes—that screens are measured diagonally.
This ambiguity is documented at least back to 1984, by IBM, the pre-eminent computer company of the time.
In 1972 IBM started selling the IBM 3333 magnetic disk drive. This product catalog [0] from 1979 shows them marketing the corresponding disks as "100 million bytes" or "200 million bytes" (3336 mdl 1 and 3336 mdl 11, respectively). By 1984, those same disks were marketed in the "IBM Input/Output Device Summary"[1] (which was intended for a customer audience) as "100MB" and "200MB"
0: (PDF page 281) "IBM 3330 DISK STORAGE" http://electronicsandbooks.com/edt/manual/Hardware/I/IBM%20w...
1: (PDF page 38, labeled page 2-7, Fig 2-4) http://electronicsandbooks.com/edt/manual/Hardware/I/IBM%20w...
Also, hats off to http://electronicsandbooks.com/ for keeping such incredible records available for the internet to browse.
-------
Edit: The below is wrong. Older experience has corrected me - there has always been ambiguity (perhaps bifurcated between CPU/OS and storage domains). "And that with such great confidence!", indeed.
-------
The article presents wishful thinking. The wish is for "kilobyte" to have one meaning. For the majority of its existence, it had only one meaning - 1024 bytes. Now it has an ambiguous meaning. People wish for an unambiguous term for 1000 bits, however that word does not exist. People also might wish that others use kibibyte any time they reference 1024 bytes, but that is also wishful thinking.
The author's wishful thinking is falsely presented as fact.
I think kilobyte was the wrong word to ever use for 1024 bytes, and I'd love to go back in time to tell computer scientists that they needed to invent a new prefix to mean "1,024" / "2^10" of something, which kilo- never meant before kilobit / kilobyte were invented. Kibi- is fine, the phonetics sound slightly silly to native English speakers, but the 'bi' indicates binary and I think that's reasonable.
I'm just not going to fool myself with wishful thinking. If, in arrogance or self-righteousness, one simply assumes that every time they see "kilobyte" it means 1,000 bytes - then they will make many, many failures. We will always have to take care to verify whether "kilobyte" means 1,000 or 1,024 bytes before implementing something which relies on that for correctness.
* Yeah, I read the article. Regardless of the IEC's noble attempt, in all my years of working with people and computers I've never heard anyone actually pronounce MiB (or write it out in full) as "mebibyte".
Why don’t kilobyte continue to mean 1024 and introduce kilodebyte to mean 1000. Byte, to me implies a binary number system, and if you want to introduce a new nomenclature to reduce confusion, give the new one a new name and let the older of more prevalent one in its domain keep the old one…
You can use `--si` for fake, 1000-byte kilobytes - trying it it seems weird that these are reported with a lowercase 'k' but 'M' and so on remain uppercase.
I gave some examples in my post https://blog.zorinaq.com/decimal-prefixes-are-more-common-th...
Agreed. For the naysayers out there, consider these problems:
* You have 1 "MB" of RAM on a 1 MHz system bus which can transfer 1 byte per clock cycle. How many seconds does it take to read the entire memory?
* You have 128 "GB" of RAM and you have an empty 128 GB SSD. Can you successfully hibernate the computer system by storing all of RAM on the SSD?
* My camera shoots 6000×4000 pixels = exactly 24 megapixels. If you assume RGB24 color (3 bytes per pixel), how many MB of RAM or disk space does it take to store one raw bitmap image matrix without headers?
The SI definitions are correct: kilo- always means a thousand, mega- always means a million, et cetera. The computer industry abused these definitions because 1000 is close to 1024, creating endless confusion. It is a idiotic act of self-harm when one "megahertz" of clock speed is not the same mega- as one "megabyte" of RAM. IEC 60027 prefixes are correct: there is no ambiguity when kibi- (Ki) is defined as 1024, and it can coexist beside kilo- meaning 1000.
The whole point of the metric system is to create universal units whose meanings don't change depending on context. Having kilo- be overloaded (like method overloading) to mean 1000 and 1024 violates this principle.
If you want to wade in the bad old world of context-dependent units, look no further than traditional measures. International mile or nautical mile? Pound avoirdupois or Troy pound? Pound-force or pound-mass? US gallon or UK gallon? US shoe size for children, women, or men? Short ton or long ton? Did you know that just a few centuries ago, every town had a different definition of a foot and pound, making trade needlessly complicated and inviting open scams and frauds?
I disagreed strongly - I think X-per-second should be decimal, to correspond to Hertz. But for quantity, binary seems better. (modern CS papers tend to use MiB, GiB etc. as abbreviations for the binary units)
Fun fact - for a long time consumer SSDs had roughly 7.37% over-provisioning, because that's what you get when you put X GB (binary) of raw flash into a box, and advertise it as X GB (decimal) of usable storage. (probably a bit less, as a few blocks of the X binary GB of flash would probably be DOA) With TLC, QLC, and SLC-mode caching in modern drives the numbers aren't as simple anymore, though.
There are probably cases where corresponding to Hz, is useful, but for most users I think 119MiB/s is more useful than 1Gbit/s.
The decimal-vs-binary discrepancy is used more as slack space to cope with the inconvenience of having to erase whole 16MB blocks at a time while allowing the host to send write commands as small as 512 bytes. Given the limited number of program/erase cycles that any flash memory cell can withstand, and the enormous performance penalty that would result from doing 16MB read-modify-write cycles for any smaller host writes, you need way more spare area than just a small multiple of the erase block size. A small portion of the spare area is also necessary to store the logical to physical address mappings, typically on the order of 1GB per 1TB when tracking allocations at 4kB granularity.
It would be nice to have a different standard for decimal vs. binary kilobytes.
But if Don Knuth thinks that the "international standard" naming for binary kilobytes is dead on arrival, who am I to argue?
https://www-cs-faculty.stanford.edu/~knuth/news99.html
Because Windows, and only Windows, shows it this way. It is official and documented: https://devblogs.microsoft.com/oldnewthing/20090611-00/?p=17...
> Explorer is just following existing practice. Everybody (to within experimental error) refers to 1024 bytes as a kilobyte, not a kibibyte. If Explorer were to switch to the term kibibyte, it would merely be showing users information in a form they cannot understand, and for what purpose? So you can feel superior because you know what that term means and other people don’t.
ls does unless you pass --si.
The author doesn’t actually answer their question, unless I missed something?
They go on to make a few more observations, and say finally only that the current different definitions are sometimes confusing, to non experts.
I don’t see much of an argument here for changing anything. Some non experts experience minor confusion about two things that are different, did I miss something bigger in this?
Approximating metric prefixing with kibi, Mibi, Gibi... is confusing because it doesn't make sense semantically. There is nothing base-10-ish about it.
I propose some naming based on shift distance, derived from the latin iterativum. https://en.wikipedia.org/wiki/Latin_numerals#Adverbial_numer...
* 2^10, the kibibyte, is a deci (shifted) byte, or just a 'deci'
* 2^20, the mibibyte, is a vici (shifted) byte, or a 'vici'
* 2^30, the gibibyte, is a trici (shifted) byte, or a 'trici'
I mean, we really only need to think in bytes for memory addressing, right? The base doesn't matter much, if we were talking exabytes, does it?
But really!?
I'll keep calling it in nice round powers of two, thank you very much.
Elsewhere you write
> They are definitely denying the importance of 2-fold partitioning in computing architectures.
No, they definitely aren't. There are no words in the article that deny anything at all.
So the "sane" options would be either not using SI for digital, or, what was chosen, change the colloquial prefixes in the digital world. The former would have been easier in the short term.