Re: OO and IBM z series COBOL was Re: Discussions of COBOLphilospphy

On Fri, 12 Oct 2007 23:35:43 +1300, "Pete Dashwood" <dashwood@removethis.ent
ernet.co.nz>
wrote:

>
>
>"Robert" <no@e.mail> wrote in message
> news:7uotg3hr7ilbu09726uatk94hm924n8d65@
4ax.com... 
>
>Possibly... I've seen very few live databases with more than a few tables,
>that were ever normalized; often redundancy is built back into them because
>some DBA thinks it is a good idea :-)

I've seen fully normalized tables at the most profitable company in the US (
ESRX). SQL
statements routinely joined 10-15 tables, and subselects were five layers de
ep.
Surprisingly, despite 500 line SQL statements,  batch jobs ran FASTER than e
quivalent
legacy code. The reason was simple: parallel processing. Cobol programs run 
on a single
CPU (with the rare exception of those that spawn threads, as I demonstrated 
here); SQL
runs on many CPUs.

>Even if we had a perfect database, there is still a case for making it
>smart. There are functions (particularly with Dates) that are much better
>suited to "back end" processing than doing them in application code.

How many languages do not have a date type? Well, there's Cobol. VB has it; 
Java and C++
have it .. I can't think of any others that don't. Should we design systems 
around
language inadequacies?

Cobol can use SQL's date functions like so:

SELECT (TO_DATE(:TO) - TO_DATE(:FROM)) INTO :DATE-DIFFERENCE FROM DUAL

On Tue, 16 Oct 2007 09:43:55 -0600, Howard Brazee <howard@brazee.net> wrote:

>On Sun, 14 Oct 2007 10:56:49 -0500, Robert <no@e.mail> wrote:
> 
>
>Does this include Unix PCs?
>
>
>
>My wife has Parallels running on her Mac, which runs a Windows VM.
>Another company has come out with one that I've read grabs a processor
>and thus runs Windows faster - as a single-processor system.
>
>I wonder what other VMs in mainframes and elsewhere do this.

Cache coherency is a hardware issue having nothing to do with VMs.
Hardwae regards all software as a single 'user'.

There are two techniques for maintaining coherency: directory and snooping/s
narfing.The
latter is faster but isn't scalable. The  writer was referring to the direct
ory technique,
which could be likened to a single-threaded lock table that knows what's in 
each CPU's
cache(s). A memory write by one CPU has to invalidate or update the word in 
other CPUs'
caches before the write to main memory can be allowed. That's  what causes t
he slowdown.

There's no need for this if CPUs share a single cache. Intel dual cores have
 a single L2
cache; AMD dual cores do not. On both brands, each CPU has its own L1 cache.