Re: Storage format
Sorry about that! Continuing: And in that case when using rows as indexes instead of columns we only need to read that specific row and might be more efficient in that case than to read a big row every time? -- Regards Erik
Re: Storage format
On Mon, Mar 1, 2010 at 4:06 PM, Erik Holstad erikhols...@gmail.com wrote: So that is kinda of what I want to do, but I want to go from a row with multiple columns to multiple rows with one column Right, and I'm trying to tell you that this is a bad idea unless you are worried about exhausting your row must fit in ram at compaction time limit.
Re: Storage format
On Mon, Mar 1, 2010 at 4:49 PM, Erik Holstad erikhols...@gmail.com wrote: Haha! Thanks. Well I'm z little bit worried about this but since the indexes are pretty small I don't think it is going to be too bad. But was mostly thinking about performance and and having the index row as a bottleneck for writing, since the partition is per row. Writing N columns to 1 row is faster than writing 1 column to N rows, even when all N are coming from different clients. Our concurrency story there is excellent.
Re: Storage format
On Mon, Mar 1, 2010 at 2:51 PM, Jonathan Ellis jbel...@gmail.com wrote: On Mon, Mar 1, 2010 at 4:49 PM, Erik Holstad erikhols...@gmail.com wrote: Haha! Thanks. Well I'm z little bit worried about this but since the indexes are pretty small I don't think it is going to be too bad. But was mostly thinking about performance and and having the index row as a bottleneck for writing, since the partition is per row. Writing N columns to 1 row is faster than writing 1 column to N rows, even when all N are coming from different clients. Our concurrency story there is excellent. That sounds good, and the same thing goes for reading, cause that is basically what I'm looking for, faster reads, not too worried about the writes. Thanks a lot!
Re: Storage format
Then you definitely want one row, range queries are slower than we'd like right now. (Ticket to fix that: https://issues.apache.org/jira/browse/CASSANDRA-821) On Mon, Mar 1, 2010 at 5:00 PM, Erik Holstad erikhols...@gmail.com wrote: On Mon, Mar 1, 2010 at 2:51 PM, Jonathan Ellis jbel...@gmail.com wrote: On Mon, Mar 1, 2010 at 4:49 PM, Erik Holstad erikhols...@gmail.com wrote: Haha! Thanks. Well I'm z little bit worried about this but since the indexes are pretty small I don't think it is going to be too bad. But was mostly thinking about performance and and having the index row as a bottleneck for writing, since the partition is per row. Writing N columns to 1 row is faster than writing 1 column to N rows, even when all N are coming from different clients. Our concurrency story there is excellent. That sounds good, and the same thing goes for reading, cause that is basically what I'm looking for, faster reads, not too worried about the writes. Thanks a lot!