Hello,

Please find some more clarifications below.

1. Matthew.Jacob >>

//
For reference in this discussion
http://physics.nist.gov/cuu/Units/binary.html
//

2. Glenn Fowler >>

//
I believe you still need to address IEEE 1541 units
    http://en.wikipedia.org/wiki/IEEE_1541
i.e.,
        5.0k  124M    # divide by 1000
        4.9Ki 118Mi   # divide by 1024
this will disambiguate displayed values and may also
help to clean up the documentation
//

Agreed Glenn.
I want to adopt this new binary prefix.
So, it would be as below.

Examples:
a> 
 /usr/bin/ksh93  'printf $"%#d\t"  5000  123456789 123456789123  
123456789123456'

  5.0k    124M    124G    124T  # divide by 1000

Only kilo(k) is in lower case.

b>
 /usr/bin/ksh93  'printf $"%#i\t"  5000  123456789 123456789123  
123456789123456'

 4.9Ki   118Mi   115Gi   112Ti  # divide by 1024

The relevant details can be documented in the man page of vmstat, similar to 
df(1M).

Copying a part from man page of df(1M)
//
 -h 

All sizes  are  scaled  to a human readable format, for  example,
14K,   234M,  2.7G,  or  3.0T.
Scaling is done by repetitively dividing by 1024.
//


3. glenn.skinner >>

//

 Date: Wed, 21 May 2008 09:34:40 -0700
    From: Eric Sultan <Eric.Sultan at sun.com>
    Subject: Re: vmstat(1m) humanisation [PSARC/2008/329 FastTrack
            timeout 05/16/2008]

    AFAIK, IEEE standards use lower-case 'k' for kilo.  This isn't, of
    course, to say that folks haven't been ignoring the standard or
    perhaps not have known it.

    Within Sun, the Editorial Style Guide conforms to established
    standards and specifies the use of lower-case 'k' for kilo.

Be that as it may, consistency with other Solaris utilities that
provide an option for requesting "humanized" output is more important.
Those utilites (or at least ls(1) and (df(1)) use upper case suffixes.

Examples to the contrary, including Linux and BSD versions of
utilities providing similar options, would be useful.

The principle of least surprise is important here.  We want the syntax
we choose to be familiar to people who have seen similar options in
other utilities.  If it turns out that there are discrepancies in this
respect between Solaris utilities and their Linux and/or BSD
counterparts, we'll have to judge what the least surprising choice is.

This principle is also important in deciding whether or not to
introduce additional suffixes to distinguish between "divide by 1024"
and "divide by 1000" units.  A wider set of examples of whether other
utilities have made such distinctions would be helpful in making this
judgement.
                -- Glenn
//

Agreed Glenn !!

I would say, after wading through the initial hiccups, the new standards
will certainly gain wide acceptance.
I would advocate the adoption of the new binary prefix standard in Solaris 
nevada.

4. john.levon >>

//
> > ** The fields can be omitted in any order.

I think you mean "emitted".
//

Fine. Will try to make it more clear as below.

**User may choose to skip fields.

5.
Apologies for the wrong usage of the word "vague" at two places.
Please find the correction for the same below.

//
*** I'm not convinced that the -h is a good idea, though.

--> The humanisation is in vague for quite some time (Example df -hl ).
Its relevant in this case as well.
The existing default output can still be obtained.
//

---> The humanisation is in usage for quite some time (Example df -hl )

//
*** I had not come across Ki, Mi, Gi,...
This avoids the major confusion over {1000/1024}

--> Looks like as of now, this convention {Ki, Mi, Gi} is still not in vague.
International Electrotechnical Commission (IEC) seems to have approved it.
//

---> Looks like as of now, this convention {Ki, Mi, Gi} is still not widely 
adopted.


Cheers,
Manjula



-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
<http://mail.opensolaris.org/pipermail/opensolaris-arc/attachments/20080522/c13c9a42/attachment.html>

Reply via email to