Wider rows may need to run through the slower 2-phase compaction process, see in_memory_compaction_limit_in_mb in the yaml file. They can also result in more GC, depending on work load etc.
Some testing I did on query performance http://thelastpickle.com/2011/07/04/Cassandra-Query-Plans/ There is no magic number. The best advice is to follow Jonathan's advice. Cheers ----------------- Aaron Morton Freelance Cassandra Developer @aaronmorton http://www.thelastpickle.com On 5 Aug 2011, at 08:22, Benoit Perroud wrote: > Thanks for your advise. Make sense. > > And without sticking to my dummy example, conceptually, what has a smaller > memory footprint : 1M rows of 1 column or 1 row with 1M columns ? > > And if the row key and column name are known, is there any performance > difference between both scenarios ? > > Thanks > > Benoit. > > > On 04. 08. 11 18:24, Jonathan Ellis wrote: >> "keep data you retrieve at the same time, in the same row."
