While I think "adaptively cache just hot parts of the row wherever
they are" is not feasible, the limitations you describe should be
fine. I think it's a clever idea, and probably widely useful.
On Mon, Jan 3, 2011 at 10:48 AM, Daniel Doubleday
<daniel.double...@gmx.net> wrote:
> Hi,
>
> we have a couple of use cases with wide rows with a small portion of hot data
> in them.
>
> Example:
>
> Chatlog:
>
> {
> $userid1-$userid2: [{timestamp: message}, {timestamp: message} ...]
> }
>
> People tend to check only the most recent pages. So while the current row
> cache doesn't work with the wide rows I was thinking of adding optional slice
> / name configs to the row cache.
>
> For example this would say: "cache the last 50 coloumns only"
>
> <ColumnFamily Name="Chat" CompareWith="BytesType"
> RowsCached="10000" RowCacheLimit="50" RowCacheReversed="true"
> />
>
> This would be limited to configurations that allow to easily determine if the
> read / write should go to the cache or not. I would try to make it as simple
> as possible by the restriction that reads can either be served entirely by
> the cache or not use it at all.
>
> I haven't found anything in jira yet and I guess this might be too
> app-specific to be generally interesting but wanted to ask if anybody out
> there already did / tried something in that direction.
>
> Cheers,
>
> Daniel Doubleday
> smeet.com, Berlin
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of Riptano, the source for professional Cassandra support
http://riptano.com
