It was an effort to increase the write throughput, but for kahadb the bottleneck is the index so any gain/loss for different workloads was over shadowed. For the moment, stick with the defaults. The logging may help identify the best value for journalMaxWriteBatchSize.
On 27 June 2013 18:55, Christian Posta <christian.po...@gmail.com> wrote: > Maybe Gary can give you more info, but the caller buffer appender is an > experimental appender to help offload the buffering of writes to calling > threads and reduce the time spent in the thread that actually writes to the > journal. The write stat window is for logging purposes to get an idea of an > avg size of write batches per WRITE_STAT_WINDOW writes. > > > > > On Thu, Jun 27, 2013 at 9:27 AM, Paul Gale <paul.n.g...@gmail.com> wrote: > >> Hi, >> >> I came across the following system properties that are uses by the >> kahaDBPersistenceAdapter, their default values are: >> >> -Dorg.apache.kahadb.journal.appender.WRITE_STAT_WINDOW=0 >> -Dorg.apache.kahadb.journal.CALLER_BUFFER_APPENDER=false >> >> A particular unit test recommends setting them to: >> >> -Dorg.apache.kahadb.journal.appender.WRITE_STAT_WINDOW=10000 >> -Dorg.apache.kahadb.journal.CALLER_BUFFER_APPENDER=true >> >> Are these type of values recommended for when one is dealing with >> persistent messages backed by slow storage? >> >> If not, can someone provide recommended scenarios when these values should >> be adjusted (per the unit test example above)? >> >> Thanks, >> Paul >> > > > > -- > *Christian Posta* > http://www.christianposta.com/blog > twitter: @christianposta -- http://redhat.com http://blog.garytully.com
