According to Ripple, at every round, validating nodes generate the ledger hash, which contains the account_hash, which is the hash of the state tree, which contains all the ledger objects, including settings and balances.
This article says that it is a very disk-intensive process, because the whole tree for the account state consists of hundreds of thousands of inner nodes which change when any of their descendants change. The amount of change depends on the depth of the affected leaf nodes.
Being so, I wonder how hard it is and much harder it will get, as the amount of nodes increase over time.
How Ripples solves this problem now and how it plans to deal with it as it grows?