On 05/07/2013 10:43 PM, Valerio Pachera wrote: > After the crash of today, my cluster is still syncing data. > From atop command I can see that network and disks I/O activity is not > so intensive: > > eth0 is generating 150Mbit/s pf traffic. > But it's a gigabit , so it could go much faster. > Disk is busy 55%. > > What do you think about? > Is sheepdog limiting the sync speed to leave I/O for guests? > Does sync have any relation with cache? Currently only one thread is responsible to recover(heal & migrate) objects. So the performance is quite bad. Multi-threaded recovery is planned. Thanks, Yuan |