I would say no. If you design around row cache and your data acceas patterns change your assertions will be invalidates and your performance may be worst over time.
I would use the kiss here. Keep it a smple usng one column family. Experiement with size teired vs leveled compaction. On Thursday, October 17, 2013, Jan Algermissen <jan.algermis...@nordsc.com> wrote: > Hi, > > my rows consist of ~70 columns each, some containing small values, some containing larger amounts of content (think "small documents"). > > My data is occasionally updated and read several times per day as complete paging through all rows. > > The updates usually affect only about 10% of the small value columns. > > Speed of the full paging is of most interest to users. > > Given the very different volatility of the per-row data, do you think my read speed would dramatically improve by splitting the less frequently changes and the very frequently changed columns into two CFs? So I can enable the row cache for the seldom changing, larger sized portion of the data? > > Or would the effect likely by rather marginal? > > Jan
