[Coco] the final stage of optimization

[Gene Heskett]

On Saturday 21 November 2009, Lothan wrote:
>If you're currently booting from a good OS-9 or NitrOS-9 boot disk/image,
>all you really need to do is merge the driver and descriptor, load the
>merged file into memory, and then init the descriptor:
>
>merge /dd/modules/rbf/rammer.dr /dd/modules/rbf/r0_128k.dd >/dd/cmds/ramr0
>attr /dd/cmds/ramr0 e pe
>load ramr0
>link rammer
>link r0
>iniz /r0
>format /r0
>
You only need to 'link' the first module of a merged group of them, which you 
did above.  And if you use 'myram', the last two steps are un-necessary since 
it formats itself on the first access, in about 100 milliseconds.  It also 
works (many times verified) at up to around 1.7 megs of ramdisk when used on 
a 2 meg machine.  Deiniz it, dmode for the size you want, and do a "dir /r0" 
and its setup and usable for the new size.  Deiniz'ing it also returns every 
byte it used to the memory pool.  Its size is determined by the sct variable 
you can change with dmode, and IIRC that value is = number of 8k blocks of 
ram it is to ask the system for and make use of.

The ramdisk, any ramdisk for that matter, being a 256 byte block oriented 
device without any formatting, really only needs the basic lsn0 (to identify 
it to os9), the FAT (up to 64k but typically only 3 or 4 sectors), the root 
directories fd.seg sector composed and written and a 64 byte root directory 
written in the next sector.  Formatting the whole capacity of a ramdisk disk 
is a huge waste of processor time.

myram I _thought_ was in the 3rdparty directory, but seems to have been 
dropped from recent (3.2.6-3.2.8) nitros9 distributions.  I only see it in an 
older /opt/os9/3rdparty directory here on this linux  machine.  But it still 
works, I have a boot disk of 3.2.8 that includes it.

Why was it dropped, Boisy / Robert?  It is good solid, bulletproof code, but 
then I'm biased since I wrote it.  :-)

Anybody that wants to play with it, I can email it to you privately.
The archive is about 7k, in lzh format.

In reading my old docs, I see that it may not set the creation date in lsn0 
correctly unless it is actually in the os9boot file, something about access 
via the F$Time function requires it to be in the system ram image for valid 
access to direct page 0 by the F$Time getstt.  The latest version also self-
adjusts to the cpu its running on, falling back to older, slightly slower 
code for the 6809's.  And slower is relative, IIRC megaread time is 11 
seconds on my 6309, and 13 seconds on a 6809.

>At this point /r0 is ready to use. Also note that I used r0_128k.dd as an
>example, but you can use either of the r0_8k.dd, r0_96k.dd, r0_128k.dd, or
>r0_192k.dd descriptors as appropriate.
>
>--------------------------------------------------
>From: "Wayne Campbell" <asa.rand at yahoo.com>
>Sent: Friday, November 20, 2009 11:45 PM
>To: "CoCoList for Color Computer Enthusiasts" <coco at maltedmedia.com>
>Subject: Re: [Coco] the final stage of optimization
>
>> When I wrote DCom I used a RAM disk. I recommended the use of a RAM disk
>> as it did speed up the program significantly. With the current setup I'm
>> using, I am unable to setup a ram drive. The NitrOS9 disk image I'm using
>> is not a proper system master, and I'm spending all of my available time
>> working on this project. I just don't have time to try to figure it out
>> right now. I'm attempting to get the copy of the OS9 system disk to boot,
>> but for reasons unknown the screen comes up black-on-black with a magenta
>> cursor, and the display commands don't change it. If I can't see what I'm
>> doing, I can't work on anything. So, I make do with what I have.
>>
>> The sort I'm using is based on the bubble sort, but instead of repeatedly
>> starting at the top and working to the bottom, it souts back upward when
>> it gets to the bottom. This way, it is constantly sorting in both
>> directions, and uses fewer iterations of the loop. It proved to be over
>> twice as fast as top-to-bottom-top-to-bottom.
>>
>> Thanks for the tips. I've heard of mergesort, but have never seen it in
>> operation or looked at its code. I know there were many things developed
>> on the mainframes for dealing with massive volume sequential data, but
>> I've never been around it. Always looking from a distance. I'll see what
>> I can find on the web.
>>
>> Wayne

[...]

-- 
Cheers, Gene
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
The NRA is offering FREE Associate memberships to anyone who wants them.
<https://www.nrahq.org/nrabonus/accept-membership.asp>

It's currently a problem of access to gigabits through punybaud.
-- J. C. R. Licklider