Subject | Re[2]: [Firebird-Architect] blobs causes table fragmentation. |
---|---|
Author | Ann W. Harrison |
Post date | 2004-10-04T21:40:16Z |
At 05:00 PM 10/4/2004, Dmitry Kuzmenko wrote:
heads. Asking BDE to SELECT * FROM TABLE is right up there with
SELECT COUNT (*) FROM LARGE_TABLE in terms of useful coding
practices. When we face a choice that will hurt a careful
program to help a stupid one, I think the world is improved by
not encouraging stupidity.
that if you asked the component nicely for the customer's name and
address, that it would return the name and address, even if the
address were a blob. If you didn't ask for the compromising
videos of the customer (a large MPEG blob), the component would
not return the contents of that blob.
The way the system currently works, the small address blob will
probably be on the same page as the name, a significant net gain.
The page array for the MPEG may be there as well, but the data
pages that hold that blob won't be loaded.
changes (availability of code, speed of processors) since that part
of the system was designed.
Regards,
Ann
>Client components don't differ is this large blob or not.Right. I'd like the system to be fast for people who use their
>Remembering BDE, using SELECT * FROM TABLE (it's way to hell,
>of course)
heads. Asking BDE to SELECT * FROM TABLE is right up there with
SELECT COUNT (*) FROM LARGE_TABLE in terms of useful coding
practices. When we face a choice that will hurt a careful
program to help a stupid one, I think the world is improved by
not encouraging stupidity.
>So, it depends on client components/driver - it load all blobsI would expect, depending on the component and the application,
>into cache or doesn't load any blobs at all.
that if you asked the component nicely for the customer's name and
address, that it would return the name and address, even if the
address were a blob. If you didn't ask for the compromising
videos of the customer (a large MPEG blob), the component would
not return the contents of that blob.
The way the system currently works, the small address blob will
probably be on the same page as the name, a significant net gain.
The page array for the MPEG may be there as well, but the data
pages that hold that blob won't be loaded.
>Why not? SQZ_block in yaffil was made on the same basis. It wasProbably I'm wrong and we should compress blobs, given all the
>made (very little code changes), tested, and produced good results.
changes (availability of code, speed of processors) since that part
of the system was designed.
Regards,
Ann