Hm... I understand that RAID0 would help to create a bigger pool for compactions. However, it might impact read performance: if I have several CF's (with their SSTables), random read requests for the CF files that are on separate disks will behave nicely - however if it's RAID0 then a random read on any file will create a random read on all of the hard disks. Correct?
-Roland 2010/4/26 Jonathan Ellis <jbel...@gmail.com> > http://wiki.apache.org/cassandra/CassandraHardware > > On Mon, Apr 26, 2010 at 1:06 PM, Edmond Lau <edm...@ooyala.com> wrote: > > Ryan - > > > > You (or maybe someone else) mentioned using RAID-0 instead of multiple > > data directories at the Cassandra hackathon as well. Could you > > explain the motivation behind that? > > > > Thanks, > > Edmond > > > > On Mon, Apr 26, 2010 at 9:53 AM, Ryan King <r...@twitter.com> wrote: > >> I would recommend using RAID-0 rather that multiple data directories. > >> > >> -ryan > >> > >> 2010/4/26 Roland Hänel <rol...@haenel.me>: > >>> I have a configuration like this: > >>> > >>> <DataFileDirectories> > >>> <DataFileDirectory>/storage01/cassandra/data</DataFileDirectory> > >>> <DataFileDirectory>/storage02/cassandra/data</DataFileDirectory> > >>> <DataFileDirectory>/storage03/cassandra/data</DataFileDirectory> > >>> </DataFileDirectories> > >>> > >>> After loading a big chunk of data into cassandra, I end up wich some > 70GB in > >>> the first directory, and only about 10GB in the second and third one. > All > >>> rows are quite small, so it's not just some big rows that contain the > >>> majority of data. > >>> > >>> Does Cassandra have the ability to 'see' the maximum available space in > >>> these directory? I'm asking myself this question since my limit is > 100GB, > >>> and the first directory is approaching this limit... > >>> > >>> And, wouldn't it be better if Cassandra tried to 'load-balance' the > files > >>> inside the directories because this will result in better (read) > performance > >>> if the directories are on different disks (which is the case for me)? > >>> > >>> Any help is appreciated. > >>> > >>> Roland > >>> > >>> > >> > > >