We are currently upgrading from ES 1.6 to 2.1.
ES is taking about 2 minutes per gigabyte of index data, according to the log timestamps.
I've turned off cluster.routing.allocation.enable and done a /_flush/synced as instructed in the upgrade docs on each node in the cluster.
This seems too slow; what should I be investigating to explain the slowness?
It depends, how much data do you have, what sort of disk, what node resources?
For each node, its about 250GB of index data and 3G memory.
The disk is a RAID using SATA drives, though I'm not as familiar with that part of the stack.
That's not a lot of resources, but maybe increasing some of these allocation settings will help - cluster.routing.allocation.node_concurrent_recoveries and indices.recovery.concurrent_streams.