Mass delete and performances

Question

Hi,
we use a few tables to log access to our various e-commerce sites.
Now some of these tables are very large (20-50 millions row) so we decided to remove old data after having consolidated it.
However deleting a large number of rows takes a long time (it took about 18 minutes to remove 2millions).
We're removing all rows with primary key less than x (key is an integer).
Does a method exists to mass delete with better performances (maybe disabling logging or ....) ?
If not, can I send a delete statement that will run on background (i.e. disconnecting the client) possibly at a lower priority ?
Rows are always inserted with primary key greater than x (it's a db sequence) so delete and insert never conflicts.
Thank you

Answer 1

Hi Andrea,
I know that this might come a bit too late for your scenario, but how about this?
For your logging table you could implement something like a poor-mans-partitioniong yourself.
Instead of one big table, you create a table for each, say quarter yourself.
You also create a synonym to he currently used "write-into-this" partition, so that your logging trigger doesn't have to be changed when the new partition should be used.
On top of that you create a UNION view over all partitions.
With that your requirements for logging everything should be fulfilled and (since you don't have to take care about updates) you can keep the active partition small and get rid of old data very fast (drop table/truncate table).
If you go for the "truncate table" (which is pretty much the same as a "drop table" for MaxDB) then you can also think about having a fixed set of tables (Q1, Q2,... Q7, Q8) that you write to in a round-robin fashion. This would then be a kind of moving-window into your logging history.
The only thing to take care of here is that you'll need to have a small time-slice in which you change the synonym every quarter.
Just my two pence on this...
regards,
Lars