Restic version: 5de36df
Observed behavior:
IO-parallelism is hard coded. E. g. defaultParallelism = 40 (checker.go:50).

The hard coded parallelism might be well chosen for solid state disks, but is not always benefical and could slow down IO-operations or cause other issues. In practice I'm observing issues checking a repository on a samba/cifs mounted network location: IO errors due to NT_STATUS_INSUFF_SERVER_RESOURCES. Reducing defaultParallelism (checker.go:50) solves these issues but requires code changes.

Expected behavior:
IO-Parallelism should be configurable with a command line parameter. E. g. restic --max-parallel-io 10 [...]. This option should be available for backup, restore and check operations.

Related: Reducing IO parallelism might also resolve #313. Exhausting resource consumption as reported in #300, #316, #298 might also be related to high IO-parallelism.