home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
kermit.columbia.edu
/
kermit.columbia.edu.tar
/
kermit.columbia.edu
/
newsgroups
/
misc.20020314-20021006
/
000049_ishikawa@yk.rim.or.jp_Sat Apr 27 10:48:43 EDT 2002.msg
< prev
next >
Wrap
Text File
|
2002-10-06
|
8KB
|
184 lines
Article: 13340 of comp.protocols.kermit.misc
Path: newsmaster.cc.columbia.edu!phl-feed.news.verio.net!iad-feed.news.verio.net!iad-peer.news.verio.net!news.verio.net!news.maxwell.syr.edu!newsfeed1.cidera.com!Cidera!newsfeed.media.kyoto-u.ac.jp!newsfeed.rim.or.jp!news.rim.or.jp!not-for-mail
From: Ishikawa <ishikawa@yk.rim.or.jp>
Newsgroups: comp.protocols.kermit.misc
Subject: Re: a bug on GNU/linux: speed reset to unintended value occasionally.
Date: Sat, 27 Apr 2002 10:49:21 +0900
Organization: Ye 'Ol Disorganized NNTPCache groupie
Lines: 160
Message-ID: <3CCA03A1.59EF10C9@yk.rim.or.jp>
References: <3CAFF81C.8039CBF8@yk.rim.or.jp> <3CC84CA6.D49F85EC@yk.rim.or.jp> <aa9l3h$4s4$1@watsol.cc.columbia.edu> <3CC98FC0.AADA5130@yk.rim.or.jp> <aac6cv$rlm$1@watsol.cc.columbia.edu>
NNTP-Posting-Host: pl940.nas911.n-yokohama.nttpc.ne.jp
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-2022-jp
Content-Transfer-Encoding: 7bit
X-Trace: news.rim.or.jp 1019872163 43612 210.139.42.172 (27 Apr 2002 01:49:23 GMT)
X-Complaints-To: root@rim.or.jp
NNTP-Posting-Date: Sat, 27 Apr 2002 01:49:23 +0000 (UTC)
X-Mailer: Mozilla 4.77 [en] (X11; U; Linux 2.4.18 i686)
X-Accept-Language: ja, en
Cache-Post-Path: duron!unknown@localhost
X-Cache: nntpcache 2.3.3 (see http://www.nntpcache.org/)
Xref: newsmaster.cc.columbia.edu comp.protocols.kermit.misc:13340
Frank da Cruz wrote:
> : I remember Fujii-san from KEK helped
> : the popularization of Kermit on Japanese PC.
> :
> Yes, the NEC PC-9801. This was in the pre-Windows days, when PCs
> ran only DOS (or very early forms of Unix). I found it quite
> amazing. The keyboard driver -- which lets you type Romaji, puts up
> little menus, learns from you, guesses which Kanji you mean, etc --
> seemed to be about 100 times more powerful than the PC itself :-)
>
Oh, those were the days. I used to get a kick out of
showing these features in one of our home-brew programs at the office
to visiting overseas dignitaries
and hearing the "Oh" and "Wow".
> Fujii-san wrote a book, "MS-Kermit Nyumon":
>
> http://www.columbia.edu/kermit/manuals.html
>
> About the PC-9801 version of MS-DOS Kermit. I suppose this is now a
> rare collector's item.
Yes, it is. The next time I see it at a second hand book store,
I will grab it right away.
> It is definitely the thinnest of all Kermit
> books. An advantage of Japanese writing: you don't have to use as
> much paper!
Until seeing the page, I didn't know books in languages other than
in English and Japanese existed. Kermit, the program, seems to
be used all over the world!
> http://www.columbia.edu/kermit/k95faq.html (third item)
Product naming is full of pitfalls. I understand.
> : Anyway, below is the result of
> : a little experiment I did to clear up
> : my understanding of some options, etc..
> : Hope some people might find this useful.
> So Cases 1 and 2 are the same, but "set control unprefix all" is a
> bit faster than "set unprefix none". OK, I believe it. Both
> commands work on the same table, but perhaps over the years the code
> for the two commands has gotten slightly out of sync. It might well
> be that prefixing or not prefixing just one certain character (such
> as NUL, ASCII 0) can make a significant difference in this case,
> since executable binaries tend contain lots of NULs. Maybe that's
> the difference between the two commands.
>
...
> I'll make a note to fix between the two unprefixing command in the
> next release.
>
Thank you for studying the data.
To give you some more data point,
I did one more experiment. [This is going to be the final test result
until probably the next year when I need to
use Kermit on a different platform and use different feature :-) ]
Based on your observation that the executable wermit may have skewed
distribution of octet values (very likely indeed), I counted the
frequency of each octet value. To make a long story short, I found
that 0 is most frequent as you guessed correctly.
In the data below, value:frequency pairs are shown.
ishikawa@duron$ ./count < wermit
0:500744 1:14331 2:7721 3:6147 4:12153 5:13718 6:6117 7:8356
******** Almost 1/4 of the total!!!
8:94806 9:4413 10:6944 11:2882 12:6407 13:3256 14:2521 15:18694
16:16293 17:3848 18:7288 19:2430 20:4744 21:4601 22:6197 23:4869
24:3524 25:2223 26:2685 27:5758 28:7488 29:8009 30:3761 31:20317
32:80939 33:6472 34:2670 35:4321 36:4350 37:7897 38:6274 39:3488
40:5673 41:3466 42:1745 43:1952 44:4746 45:4373 46:6375 47:2859
48:4359 49:4954 50:2970 51:1510 52:2852 53:7989 54:1523 55:1883
...
Of the approximately 2MB ( 2183313 bytes to be exact), close to
one fourth (1/4) is byte with value 0!
So again, I got curious and created a gzipped file from wermit.
cp wermit /tmp/wermit
ishikawa@duron$ gzip --best -v /tmp/wermit
/tmp/wermit: 63.3% -- replaced with /tmp/wermit.gz
ishikawa@duron$ ls -l /tmp/wermit*
-rwxr-xr-x 1 ishikawa users 800501 Apr 27 10:26 /tmp/wermit.gz*
Why compression?
Since compressed file usually have more balanced, close to uniform
byte value distribution, I tried sending the compressed file and
measured CPS to see if the uniform distirbution
closes the gap between observed CPS values in case-3 and case-4.
(The uniformity is like this. It is indeed very good.
ishikawa@duron$ ./count < /tmp/wermit.gz
0:2903 1:3066 2:2973 3:3186 4:2985 5:2881 6:3127 7:3284
8:2993 9:3010 10:2990 11:2996 12:2944 13:3282 14:3016 15:3456
16:2729 17:2907 18:2784 19:3009 20:2820 21:2889 22:3073 23:3147
24:2945 25:3013 26:3033 27:3129 28:2889 29:3015 30:3113 31:3700
32:2844 33:2866 34:2722 35:3060 36:2838 37:2790 38:2891 39:3220
40:2948 41:2901 42:2798 43:3017 44:3006 45:3199 46:2923 47:3343
...
The new CPS values are shown below
appended to the original table.
>CPS (measured into 20% of the transfer. The file is 2183313 bytes.)
>============================================================================
>CPS: case-1 case-2 case-3 case-4
>---------------------------------------------------------------------------
>send: 3010 3009 3835 3621
>receive: 2973 2972 3786 3568
CPS measured for compressed wermit file. (wermit.gz)
I only measurement for case-3 and case-4.
(Also, the measurement was done into 50% (approximately 400KB)
of transfer.)
case-3 case-4
3796 3721
3753 3702
===========================================================================
So as you have again correctly guessed, the skewed distribution of
octet values (0 being dominant) had something to with the CPS under
different control/prefix setting, i.e., "set control unprefixed all"
vs "set prefixing none".
With the uniform distribution of octet values, the difference
decreases very much. (So maybe the code path is indeed similar(?) and
has not diverged very much??? The slight difference about 50 could be
indeed a statistical fluke, but... )
Hope this helps.
Thank you again for the great support.
Always hoping to send back feedback to improve KERMIT.
Ishikawa, Chiaki
PS:
>case-4: `case-1` setting
> + "set prefixing none"
> (BTW, Completion doesn't show "none" as a valid third argument.)
I noticed a few cases where an accepted option is not shown in the
response to "?" completion helper listing. The above is such a case.
I found the effect of "minimal" and "none" is slightly
different and so "none" ought to be listed IMHO.
To wit,
(/home/ishikawa/KERMIT/new-kermit/receive/) C-Kermit>set prefixing ?
control-character prefixing option, one of the following:
all cautious minimal
(/home/ishikawa/KERMIT/new-kermit/receive/) C-Kermit>set prefixing