Shridhar Daithankar wrote:
> Perhaps you could attempt to store a fix small number of records per row, say
> 4-6? Or may be a smaller fixed size array, That should make the row overhead
> less intrusive...
Thanks, I didn't like your idea, but it helped me come up with another
idea:
(timeseries_id integer, top text, middle text, bottom text);
The entire timeseries is the concatenation of 'top' (a few records),
'middle' (millions of records), and 'bottom' (a few records). To get
the last record, or to append a record, you only read/write 'bottom',
which is very fast. Whenever the entire timeseries is written (a less
frequent operation), the division into these three parts will be
redone, thus keeping 'bottom' small.