[c-nsp] Bit, byte huh?

[Richard A Steenbergen]

On Wed, Jun 11, 2008 at 10:31:44AM +0300, Ziv Leyes wrote:
> One of the first things I've learnt is that bandwidth is measured in bits while memory/disk space/ file size are measured in bytes.
> Since then I always had arguments with my colleagues, for example, when I worked at a very large ISP, the customer's links in bigger circuits were limited by traffic shaping the sub interfaces, so the guys that "taught" me how to do it told me to set it like this, for example of a 4/2Mb link.
> 
> interface atm0/1.30
> bandwidth 4096 (this was only for the MRTG)
> rate-limit input 4096000 blah blah blah...
> rate-limit output 2048000 blah blah...
> 
> 
> And I always told them that we sell BITS, not bytes, but the "marketing" convention got the customers used to talk in bytes, perhaps IE browser is guilty because when a customer measures his bandwidth it checks at how many BYTES per second they download a file from a website. Therefore, when you sell 4Mb you need to give them 1x1024 bandwidth units, 1M is 1024, 2M is 2048, and so on.
> I gave up on this battle, I can't change the world! But I still believe that 1M is one million BITS!

The distinction isn't so much about bits vs bytes, it's how the data is 
stored and accessed. The only place you see things talked about as 
power-of-2 is inside a computer when you are accessing memory (and thus by 
association, when you are talking about the size of files stored in memory 
as well), because what you're actually doing is addressing 10 bits of 
memory to get 2^10 = 1024. Some lazy folks out there noticed that 1024 was 
"pretty close" to 1000, so when they went to access a 1024 byte chunk of 
memory they called it a kilobyte, and generations of confusion and 
inaccuracy ensued. The proper name for this is a kibibyte (1 KiB = 1024, 
MiB for mebibyte, etc), and it is NOT the same as a kilobyte, nor is every 
reference inside (or associated with) a computer going to be 1024 based.

Bits across a telecommunications wire are in no uncertain terms measured 
in real SI units, not power-of-2. As you move up to scale from kilo to 
mega to giga the inaccuracy increases by 2.4%, so by the time you hit 
megabits you're off by 4.8%, and gigabits are off by 7.2%. I like to say 
that people who don't understand this are forever doomed to pay a 4.8% tax 
for being bad at math. Sadly, I can actually think of at least one tier 1 
network who doesn't understand this, and their bills are always 
mysteriously much lower than they should be. :)

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)