[Gllug] Oracle on Linux

Nix nix at esperi.org.uk
Sat Sep 18 21:49:22 UTC 2004


On Fri, 17 Sep 2004, Richard Jones stated:
> It turned out that some clever Informix drone had added the equivalent
> of -Dvolatile= to the compilation flags in order to "get it to
> compile".  This caused the compiler to optimize out any references to
> shared variables.  The inter-thread locking was just broken.  We went
> the expensive route and switched to Solaris after that.

That's... impressively dumb. Probably the poor sod had been told by her
PHB `it *must* be out the door by {date}, working or not' and had taken
him at his word...

> Limited size strings.  In fact all sorts of arbitrary limits all over
> the place.

Informix, like Oracle, took the Unix Way of `stupid arbitrary limits'
to heart. :(

(Oracle plugged as a *feature* in Oracle Server 9.1 its ceasing to
mlock() its entire address space and its understanding of this new
`virtual memory' thingy.)

>             Like tablespaces - if you forgot to allocate exactly the
> right block size and lay out your tablespaces correctly, then your
> tables could unexpectedly hit some limits, even when you had plenty of
> disk space left.  Oh, and you had to remember to do this when you
> created the table, because there was bugger all way to adjust it
> afterwards.

A failing Oracle shares :((((

> A database access tool called 'dbaccess' that sucked even in
> comparison to sqlplus.  For one thing, it had an 80 column limit.  You
> simply couldn't type past the right hand end of the screen.

Shades of the C64. (And a sign of a fixed-size buffer in the terminal
handling code.)

> Slow.  Very very slow when compared to PostgreSQL.  Often you'd have a
> perfectly good SQL statement, and make a minor tweak to it, and the
> statement would run 1000s of times slower.

Query optimizers? What are they?

> Weird debugging tool called 'onstat'.  It had about a million options,
> and allowed you to enumerate almost any aspect of the database, which
> was good, but the syntax it used was just terrible.  Instead of
> printing out table names, for instance, it printed out hex numbers
> which you had to break apart manually into binary and cross-reference
> each bitstring through several other displays to work out what table
> it was talking about.  Computers are supposed to do this shit, not me!

Oracle has much the same, enabled in a different way, and totally
undocumented. (In fact they've gone so far as to threaten to sue people
who document these switches... at the same time as their own support
people are pointing enquirers to exactly the same webpages. Right hand,
meet left hand.)

-- 
`The copyright file is for everyone.  That we make it available in
 plain-text, uncompressed form rather than in spinning, throbbing
 OpenGL-rendered 3D text over a thumping dance music soundtrack is a
 feature, not a bug.' --- Branden Robinson

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list