cql3 - Cassandra Performance : Less rows with more columns vs more rows with less columns -
we evaluating if can migrate sql server cassandra olap. per internal storage structure can have wide rows. need access data date. need access data within date range have financial data. if use date partition key support filter date,we end having less row huge number of columns. hamper performance if have millions of columns single row key in future process millions of transactions every day.
do need have changes in access pattern have more rows less number of columns per row.
need performance insight proceed in either direction
using wide rows typically fine cassandra, there few things consider:
- ensure don't reach 2 billion column limit in case
- the whole wide row stored on same node: needs fit on disk. also, if have dates accessed more other dates (e.g. today) can create hotspots on node stores data day.
- very wide rows can affect performance however: aaron morton last pickle has interesting article this: http://thelastpickle.com/blog/2011/07/04/cassandra-query-plans.html old, believe concepts still valid.
for table design decision 1 needs know typical filter conditions. if have other fields typically filter exact match, add them partition key well.
Comments
Post a Comment