问题
I am developing a system that will have a database backend. I am intending that every table have a PK that is arbitary, system generated and maintained.
My environment (for development and production) is Windows 7; Delphi; and an embedded database (probably Firebird).
My data structures include one table OWNER, that has very little information associated with it - probably no more than name and description. The rest of the data structure will be about 50 tables, each with name and description, together with a list of other attributes - say 20 per table on average. In addition there could be 20 associative tables most of which will have only FK attributes, though a small proportion will have additional attributes. It is my intention (at least to start with) to have the schema fully normalised.
For a given OWNER, most table will have O(10^3 to 10^4) records, though one or two will have O(10^5 to 10^6). The number of OWNERs is likely to be O(10^2 to 10^3). Most accesses are likely to be clustered - there will be a substantial number of accessess for one OWNER, and then a substantial number of accesses for another OWNER.
Every data item will belong to exactly one OWNER, and cannot be transferred from OWNER to OWNER. All accesses to the database will know what OWNER they are using; no access will ever have to access the contents of more than one OWNER.
I am aware of the following three options in designing my schema:
- Use the OWNER table as described. Use the OWNER PK as a foreign key in every table (though possibly not the associative tables). Add the OWNER PK as an additional clause into every query, join, stored procedure, view etc.
- Add a column to the OWNER table containing a small code - say a four digit integer. For each OWNER, create a duplicate set of tables - the table names for a particular OWNER will have the appropriate code added as a suffix. This would require every access to have previously obtained the suffix code from the database. Then the access will be of the relevant set of tables.
- For each OWNER, create a duplicated database, with the tables and table names being duplicated. This implies that there would probably be a subsidiary common database containing data about the relevant duplicate database. Again this common database would need accessing before any access - or series of accesses for one particular OWNER.
What are the pros and cons of these different approaches? Have I missed any other options for overall design?
Edit reversed - I provided my own answer instead.
回答1:
Option #1 is by far the best. Obviously you are planning to use a system generated id number for this record as well...otherwise you will run into a problem if your OWNER pk is the person's name and you end up with multiple John Smith.
Option #2 would work fine, but adds too much complexity. So long as you already have the OwnerId as the PK in OWNER table, jsut fk that for the rest and save the additional overhead and effort.
Option #3. This option is great if you have a poor front-end interface and want to be able to easily read through the records manually and make sense of them...but all the extra tables will slow you down and create extra sub-routines that have to be written into the front-end to keep track of the tables.
回答2:
Continuing to think about this, in the light of @techtheatre's response, I realised that any changes to the schema would imply running change scripts a 1000 times for options 2 or 3. This is obviously not sensible.
来源：https://stackoverflow.com/questions/4911658/pros-and-cons-of-different-schema-designs-for-tables-that-all-have-a-relationshi