🚀 go-pugleaf

In article <18200004@gorgo.UUCP>, bsteve@gorgo.UUCP writes:
> 
> To off unbiased commentary on this statement; I have observed that in
> comparison tests on identically populated databases on the same machine,
> Oracle access times are very similar to those of Informix SQL. This does
> not take into account the percieved much greater speed afforded by Oracle's
> ability to lookahead on searches. This comparison was also not conducted
> using raw devices which I am led to believe can lead to as much as a 30
> percent improvement. There was one VERY significant area in which Oracle
> was a FULL ORDER OF MAGNITUDE faster than Informix SQL, on nested selects.

- Most of the UNIX RDBMS (Oracle,Ingres,Unify,Informix) are getting to be
  good enough that the retrieval speeds are similar for small amounts of
  data in simple queries (after all thats what the indexes are for :-)).
- The ability to do asyncronous read-ahead wins big in a "live" database
  where B-tree reoranization has caused data and index blocks to get
  scattered all over the disk.
- There are many points to using raw devices:
   - Faster I/O (buffers are read directly into the user's data space)
   - No overhead for UNIX indirect blocks.
   - Guaranteed data integrity for systems not supporting UNIX Buffer Cache
     write through.
   - Much decreased impact of your RDBMS on other UNIX applications (when
     it flushes your buffer cache by filling it up with blocks from the d.b.)

I wonder if the results would have been the same if:
  - There had been controlled measurement of impact of the RDBMS on UNIX
    applications or UNIX applications on the RDBMS.
  - Efforts had been made to "jumble" the d.b. so they would look more like
    a production d.b.

-- 
Robert Bradbury
Oracle Corporation
(206) 364-1442                            {ihnp4!muuxl,hplabs}!oracle!bradbury