You should consider dropping the index before the insert and reinstating it (aka "rebuilding the index") after the mass-insert is done. It seems a bit odd that you would use an "Android Machine" for working with large amounts of data. You might be exposing file system limitations. Is it the case that your insert data is already ordered in a way that forces b-tree rebalancing disproportionate to the log of the record count? If that is so, a different insert order could help. And if it is really true that dropping the index only "helps a little", then something strange is happening, such as excessive rebalancing or something else not evident in your problem statement.