Subject Re: [Firebird-Architect] Feature Request
Author Christian Stengel
Hi Jim,

> The way to do a fast load is to load the data with indexes disabled
> then
> enable the indexes. The indexes will be built using a "fast load"
> scheme from a sorted data stream, which is much faster than incremental
> index updates. This does not require additional database support.

Will this be faster than in firebird? I am doing that currently this

1.000.000 records take:
No index (1. 1.000.000 records): 0m46.865s
With 1 index: 1m13.135s

Unfortunately - the more data you get in the longer it takes (for the
10. 1.000.000 records it takes 1m12.937s) - without any indices.

creation of index with 10.000.000 records: 3m16.072s - so yes. I am
saving time.

>> 3. metric calculations
>> Ann sometimes asked whether there is a need for spartial extensions
> That's what UDFs are for. Use them.

OK. But internal functions would be nicer - easier to handle, you can
be shure, that they are available on all platforms ...

>> 4. compressed data field
> Recordds are already run-length compressed. The zip compression
> algorithm loses big time for short items. A 6 character string, for
> example, bloats in 118 bytes. The algorithm is space efficient only
> for
> large strings.

Great. In FB 1.5.2 I can edit a .gdb file with vi and see the strings

>> 5. Index file for external tables
> Not a chance in the world unless the external file manager d

Can the filemanager easily be exchanged in Vulcan (e.g. without the
need of recompiling the whole engine?).

> oes it
> itself or you wish to write your own external file manager. External
> files are useful to import or export data or to interact with an

External files are slower than importing data with c-api in current
versions of firebird. The only benifit is, that it's faster to
implement them. ( 0m46.865s vs. 0m55.354s with external tables)

> application that uses RMS. It is not intended to every be an
> alternative record storage format.

Yes. I know that. But when you have to deal with 35.000.000 records per
day from one source - you have to be creative - and one aproach could

Use external files as temporary storage, agregate the data and throw
them away after a month or so :-). And in that case - a one time index
could save time :-)