Dear All,
I have 2 data dumps (comma separated) each with around 53,000 records (
just sample data. it could be 10times more than this in real time).
I need to write a script to -
1. find matching records from these 2 dumps based on a set of matching
fields
2. store matching records from each dump into database
3. find the remaining records from each dump
4. find matching records by excluding one of the matching field
5. again store matching records from each dump into database
For step 1 I used "cogroup"
For step 3 I split "cogroup" with nulls for dumps 2 & 1 respectively to
get the remaining records for dumps 1 & 2
For step 2 & 4 I used DBStorage UDF to store the records into DB. With
this approach I get 4 store commands (2 commands for each dump at steps
2 & 5).
Before storing to DB I'm using another UDF to generate a running
sequence number which will be stored as key for each record being stored.
Problem:
=====
The script for this entire process is creating 6 map-reduce jobs and
taking about 10mins to complete on a cluster of 3 machines (1 master and
2 slaves).
The same requirement when done using a stored procedure is completing in
5 mins. Now I'm worried that my script could kill in real time environments.
Requesting to suggest -
-> What am I missing?
-> What can I do more to improve the performance that is in comparison
to stored procedure?
-> What changes and/or additions to be done so that the script is
scalable to any amounts of data?
Thanks in advance,
Regards,
Sarath.
