SSTableLoader performance limits during large-scale snapshot restoration

Apache Cassandra utilizes the sstableloader utility to stream snapshot data back into a live cluster, ensuring that data is distributed according to the current token ring. This process is critical for recovering data from point-in-time backups across multiple nodes.

When restoring massive datasets, the streaming process introduces significant network and CPU overhead. There is a design tension between the speed of data ingestion and the stability of the target cluster's read/write availability.

What are the documented mechanisms for throttling sstableloader to prevent cluster instability during restoration? Are there specific configuration limits that determine the maximum concurrent streams allowed per node to avoid saturating the network?