Re: Multiple keyspace question

2012-07-10 Thread Edward Capriolo
A problem of many keyspaces is clients are bound to a keyspace so connection pooling multiple keyspaces is an issue. Cql has support for some limited cross keyspace operations. On Sunday, July 8, 2012, aaron morton wrote: > I would do a test to see the latency difference under load between having

Re: Multiple keyspace question

2012-07-08 Thread aaron morton
I would do a test to see the latency difference under load between having 1 KS with 5 CF's and 50 KS with 5 CF's. Your test will need to read and write to all the CF's. Having many CF's may result in more frequent memtables flushes. (Personally it's not an approach I would take.) Cheers --

Re: Multiple keyspace question

2012-07-06 Thread Shahryar Sedghi
Aaron I am going to have many (over 50 eventually) keyspaces with limited number of CFs (5-6) do you think this one can cause a problem too. Thanks On Fri, Jul 6, 2012 at 2:28 PM, aaron morton wrote: > Also, all CF's in the same KS share one commit log. So all writes for the > row row key, acro

Re: Multiple keyspace question

2012-07-06 Thread aaron morton
Also, all CF's in the same KS share one commit log. So all writes for the row row key, across all CF's, are committed at the same time. Some other settings, such as caches in 1.1, are machine wide. If you have a small KS for something like app config, I'd say go with whatever feels right. If

Re: Multiple keyspace question

2012-07-06 Thread Robin Verlangen
Hi Ben, The amount of keyspaces is not the problem: the amount of column families is. Each column family adds a certain amount of memory usage to the system. You can cope with this by adding memory or using generic column families that store different types of data. With kind regards, Robin Verl