You can use metasave to check the bottleneck of decommion speed, If the bottleneck is the speed of namenode dispatch. You can tuning dfs.max-repl-streams to a large number (default 2). If there're many timeout block replication tasks from pending replication queue to need replication , you can tuning dfs.replication.pending.timeout.sec to a smaller numer, to make block replcation more positive.
Pay attention!! Please check your hadoop version, if block transfer has no speed limit, the bandwidth may be stuff full Thanks & Best regards Baggio 2010/12/16 sravankumar <[email protected]> > Hi, > > > > Does any one know how to speed up datanode decommissioning and > what are all the configurations > > related to the decommissioning. > > How to Speed Up Data Transfer from the Datanode getting > decommissioned. > > > > Thanks & Regards, > > Sravan kumar. > >
