HP3000-L Archives

March 2003, Week 1

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Wirt Atmar <[log in to unmask]>
Reply To:
Date:
Mon, 3 Mar 2003 13:37:48 EST
Content-Type:
text/plain
Parts/Attachments:
text/plain (39 lines)
Patrick writes:

> The primary path designation was originally intended for DBUNLOAD/DBLOAD,
>  now utilities will offer it as a default in certain operations (Adager and
>  its Repack command, for instance)
>
>  A good rule of thumb would be to deisignate a primary path as the path that
>  your users access the most and has a > 1 average chain length (for sets
>  that have > 1 path).

In the late 1970's through the middle 1980's, we used to use the primary path
designation in QueryCalc's search optimization decisions -- if the user had
specified a query question with multiple search items in it. It had real
value then. Because you can only search down one chain at a time, you would
like to insure before you start that it's the shortest chain. In a
"late-binding" optimization operation, QueryCalc measures the length of each
search chain and chooses the shortest one for its search path. In earlier
versions of QueryCalc, when making that chain length comparison, we also gave
an advantage to the primary chain, artificially "shortening" its length by a
fixed 30%, simply because on older systems, the primary chain's records were
likely to have better "locality" on the disc (more of its records would be
next to one another as compared to a non-primary chain's records).

There's no way that we could really know on any specific search if the
primary chain had a 30% advantage; actually calculating that information
would have been intolerably slow. Rather, it was simply an average number we
derived in the middle 1980's regarding search responses on the last of the
classic machines. However, as time progressed and the users' machines became
increasingly more memory rich, and we didn't have to access disc nearly as
often, and MPE's memory management in the RISC machines became so much
better, the advantage of the primary chain evaporated with evolution of
MPE/iX, thus we no longer give it any advantage nowadays over any of the
other chains.

Wirt Atmar

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *

ATOM RSS1 RSS2