Re: Slashdiot article about P6

On 1/8/07, Simon Coulter <shc@xxxxxxxxxxxxxxxxx> wrote:

I should know better but this append is such complete rubbish I can't
help myself. It's not that I expect Steve Richter to change his views
but rather that someone else might believe the tripe he spouts.

On 08/01/2007, at 12:36 AM, Steve Richter wrote:

> there are so many. i5/OS has not been improved since the early 1990s.

Have you been in a coma since then? You're saying there have been no
improvements for over 15 years? That's stupid! Let's see:

1992 OS/400 VRM220 announced:
       - Save/Restore while active
       - DB expanded space, key size, and access path size

great, I am not going to quible with what you consider an improvement,
only, for the programmer, there has not been much since v5r1 in 2000.
Both feature wise and CPW horsepower.  I have a 170 purchased from IBM
via ebay. The system has 460 CPW and v5r1.  This system is 6 yrs old
model wise, yet today an i520 with v5r4 gets me 600 CPW, good stuff
added to sql procedures and a few more RPG bifs.

And this is only the big ticket stuff. In addition, each release has
supported additional hardware and software, compiler enhancements, plus
new OS/400 commands, system values, and job attributes. "Not been
improved since the early 1990s"? Don't think so! Just more proof you're
talking rubbish.

no, it is proof my focus is programming the system.

>
> long object and field names

There has been discussion on this list previously about the
difficulties involved in doing this for native objects. It will cost a
LOT of money to rectify for almost no gain. Only programmers care about
the object name limitations; end-users and managers (i.e., those who
make the business decision to buy the system) don't give a toss. No
return on the investment so IBM are unlikely to do this.

right, only programmers care. Big mistake of IBM's to devalue programmers.

You seem to want IBM to spend money on development efforts that will
make you happy but no-one in a position to influence the purchase
process gives a flying f__k about.

I hear you loud and clear. What I fear is IBMers echo your sentiments.
The as400 was great because it was a premier system for admins AND
programmers.

Just 'cause you want to run Oracle or MySQL on the box doesn't indicate
demand from other sources. If Oracle want to port their DB to i5 they
could do it (with or without IBM's help). Wouldn't be much point though
because DB2 craps all over Oracle in terms of less administration and
increased limits.

this is academic, but I was reading recently how .NET 2.0 was improved
so that SQL server could participate fully in the managed code world.
The author wrote that SQL Server like all databases needs to manage
its own main store memory allocations for optimal performance.  I have
always been interested in how database access performs so well on the
as400 compared to the difficulty I can have getting my code to run
well.  The point is a database port is not going to perform as well as
db2 because db2 is most likely wrapped tight with the paging
mechanisms of the system.

Refer to the many options of SETACST for an idea of what it takes to
tune a memory bound application.:

http://groups.google.com/group/comp.sys.ibm.as400.misc/browse_frm/thread/e9809f86998c2568/9d9288164a0c0313?lnk=gst&q=setacst&rnum=1&hl=en#9d9288164a0c0313

If you want MySQL or some other kiddie database then port it yourself.

I would like to do that. In ILE C instead of PASE resident GNU C. Need
full p5 speed to make it worthwhile.

>
> Scrap spooled files. replace with XML documents.

Two completely different things: Spooled files are printer-ready data
streams, XML documents are just text with tags. That'll look nice on a
printout. No printer currently transforms XML into a formatted report
so how would these XML documents be printed? By processing them in a
program that generates a printer-ready data stream. Hmm, isn't that
where we came in?

Maybe you want to use DDS to generate XML instead of a printer file?
Why, it's very easy for a programmer to generate XML, quite a bit
harder for the same programmer to generate a printer-ready data stream
(level of difficulty depends on the target printer). Why require IBM to
expend development dollars on something easily within the capabilities
of all current programmers?

As far as I know, no other OS generates XML for printing or uses XML to
store database records. Why? Because there are no proper standards for
such a thing. So which dialect of XML should be adopted to describe
printed output?

the bottom line is output queues, as invented by G Glenn Henry's
group, way back in the begin times, are still a terrific way to
organize output.  It is IBM's lock on the format of that output that
makes spooled files archaic and of little value to the network. If the
spooled files were in XML format you could easily write drivers that
would "print" the output to a printer, excel, another database.

>
> expand pointer size so the 16meg limit on spaces and strings can be
> eliminated.

Need more than 16MB? Use teraspace. IBM have provided a solution to the
size limit, all you have to do is use it. Besides, it's not the pointer
size that imposes the limit but rather the segment-based addressing
mechanism which has nothing to do with SLS but rather is a side-effect
of the object-based design.

this makes no sense. objects are things with interfaces and no
implementation details from the perspective of the application using
the object. So a user space object would just be a stream of bytes
from the perspective of the application writing to it. No limits to
its size, no need to flush its contents to disk. It is the segment
size limitations of the SLS that mucks up the usage of the user space
and forces the programmer to deal with the implementation details of
the user space object.

> or scrap/redo the single level store architecture.
> Possibly, limitations and security problems of the SLS is what is
> preventing IBM from investing in our system.

Where do you get that idea? Smacks of 2+2=5 (which can be true for
sufficiently high values of 2 but otherwise is patently false).

> You dont need the SLS to get the features of i5/OS.

SLS is one of the fundamental architectural principles of i5/OS. Bit
hard to remove that and still say it's i5/OS (or OS/400). Next you'll
want to remove object based design and technology independence. You
need to do some research on exactly what SLS is and isn't because I
think this is yet another thing you don't understand.

give me an example of what SLS has to do with as400 objects?
Obviously a database does not need SLS to run. The materialization and
encapsulation of programs could be accomplished with stream files on
the IFS.

>
> better support for multiple signatures in service programs. Previous
> signatures should auto map the export number of srvpgm exports from
> what they were when the signature was current to what they are in the
> current signature.

Why? That would stop me from renaming an old function and still have it
continue to work with programs using an older signature. Your approach
would force me to use new names for any replacement function instead of
renaming the OLD function. The new function would also have to
distinguish between being called the old way and the new way.

What IBM have given us with signatures is better than your suggestion.
The only thing missing is procedure signatures which would give us
procedure overloading and probably satisfy what you're asking for.

Maybe you don't really understand how service program signatures work?
Check the archives--signatures have been fully discussed previously.

wow, you are ticked off. I suspect a detailed response on my ideas for
how signatures and srvpgms could be improved would be a waste of time.
The bottom line is apps should not get signature violations and
binding source should not be necessary.  Using export numbers alone as
the means, at run time, of finding the procedure in the srvpgm, is a
relic meant for the day when CPU cycles were slow and expensive.

>
> ability to create associated spaces of an object.  did you know a
> srcmbr can have an associated space? could use the associated space of
> a srcmbr to store meta data such as extended member text, or info used
> by a precompiler.

Open a DCR and request this feature. IBM have provided APIs for working
with an associated space of a program object. If you can place a
business case for doing the same with other objects they'll probably do
it.

It might be nice to use an associated space for the purpose you
describe above, and there are distinct advantages to so doing, but it's
not necessary. You can achieve the same result via different means.

staying with the theme of the low importance of the programmer to the
success of the as400, I see what you mean.

> a garbage collection memory model that holds objects. this means you
> need a facility for defining and storing object types. ( you need a
> reflection type facility for calling the dispose method of the object.
> )   One use of something like this would be system APIs would return
> objects that are typed, that are self describing. In the case of a
> list objects API, the system would return an object that is holds an
> indexed array with the capacity to list all the objects on the system.

What would this achieve over the existing list APIs? I can see possible
benefits in a different programming approach but this really adds no
commercial value. Again, this would cost LOTS of money to do and have
limited return to IBM. If you see otherwise then write the business
case. Show us and IBM how this would increase sales, revenue, and
profit from i5. That means explaining why those with purchasing power
would buy the system--not why programmers looking for a cheap or free
development system would be interested. They've got no money and no
power so they're not important.

what can be said here. .NET and Java are very successful because of
their GC memory models.  My comment on the lack of threadsafeness of
i5/OS being a detriment to Java being a full player in i5/OS has not
been replied to.  Modern systems need a modern GC memory model. They
need a lot of CPU to support feature rich object style programming.

>
> etc, etc. most important, if the system was able to run at p5 speeds
> and prices, 3rd parties could provide a lot of what is needed to
> improve the OS and language support.

This drivel again. The big block to 3rd-parties providing extensions to
OS and compiler languages is the restricted access to the guts of the
system. However, in most cases that's not necessary. There are plenty
of garbage collection based, scripting languages that can run like a
rocket on this system and they don't need access to system internals to
do it. Lua is one of them, Ruby is another. Don't need so-called p5
speeds to run them either.

great. I would like to know more. Never heard of Lua. Any web links to
info on Ruby apps running on the i5?  especially a geared down i5?

ILE gives you the tools to extend the current programming languages.
Garbage collection is not required for that but you could build such a
thing if you wanted.

The OS can be extended with user-defined commands and processing
programs.

Many system objects are just variations of space objects and index
objects. You could build your own object types from user space and user
index objects. Sure, you wouldn't be able to define an external object
type for them but you could accomplish your goal.

So if you want to improve the OS you have suitable tools--maybe not the
ones you'd like but perfectly acceptable and usable. Instead of
whinging about what IBM are failing to do technically why don't you
create the stuff you consider missing and make it available?

working on it.  The pre compiler design of the system is very good.
Esp the way debugger views can be linked to precompiler source code. I
think there is a lot of potential on that front, adding features to
ILE C and RPG.

I've seen some of your code. It suggests you're a good programmer but
your comments regarding IBM's development and suggestions for what must
be done to improve i5/OS just make you seem a fool.

-Steve