VARRAYS for VLDB

Hi,
 I am about to decide on the physical datamodel for a rather large database,
 I would appreciate some feedback before it is all messed up.

 There is a number of measures, that are sampled with a certain frequency.
 i.e. temperature in centigrades @15 samples per second. I get several hundred/ thousand of these streams, and the data is to be kept for at least three months.

 The interesting part is the table that stores the samples:  Partitioned by timestamp, subpartitioned by sample type, or something like that.
 This might even become a index-organized table. Not sure about that yet, I will see how the prototype works out.  This is to hold several billion rows.  

 To cut down on the overhead for primary key values etc., I was thinking about using a varray to hold the actual sample values - so i can get maybe 50 samples per row in that table.  

 Then in the table, i can keep several statistics such as number of samples, average, etc.

 The whole intend is to get the storage requirements into a reasonable range (i.e. less than a terabyte) for that table.

 To get the logical view back, I would create a view that rolls out that varray,
 so the application sees indeed one 'pseudo-' row per sample.

 Is this reasonable, or is there any showstopper with varrays that you know about ?

Thanks for your help
 Karsten Received on Wed Apr 23 2003 - 07:03:08 CDT