Message-Id: <9403030912.AA03171@us2rmc.bb.dec.com>
Date: Thu, 3 Mar 94 04:12:48 EST
From: touche!  03-Mar-1994 0405 <groff@epee.enet.dec.com>
To: dagoura@MIT.EDU
Cc: groff@epee.enet.dec.com
Apparently-To: dagoura@mit.edu
Subject: RE: re: Plea 2

way cool

I am running out of steam... but I basically have the db schema mapped out.  

I still haven't got the indices defined.  I "kind-of" want to normalize the
tables too much and am looking at denormalization now.  I think I just need to
combine two tables and the indices fall out just fine.  And I wonder what other
constraint I will put on the DB.  There will be some "mandatory" fields and
some "default values"... so all-in-all we are close.

Next is the deliver deamon code... that is the nasty stuff.  I will probably
parse the incoming messages with a C++ program and then call into the database
(this is for "speed" and recoverability).  I am also worried about "caseless"
searches.  Right now:  "Danulf" != "danulf".  I could write a constraint to
deal with this... but that might be a v1.1 enhancement.

[maybe I should just sleep :-)]

I am also tweaking the data structures to be all "quadword" aligned.  That will
be usefull when EPEE becomes an AXP machine.

I have developed a scheme for queue requests on this machine and then queuing
"failed updates" for later "hand" processing.  Its no major trouble... but I
had to consider "how" to do it.  There are HOLES everywhere that a REAL TP
system would just die over... if we have a problem, I can patch in ACMS (A real
TP system) with little effort.  

BTW: this also allows me to take the db down for maintenance and not lose any
updates.

But, I was wrong...  its not a 1 hour implementation... but the weekend should
be enough... of course I am adding complications... oh well... :-)

then to testing :-(

-Danulf