Can someone explain what exact latency is measured by biolatency tool (bcc)?

Question

Is it:

1. Time between the Virtual FileSystem calling the block IO layer to the time it submitting requests to the block device driver?
2. Time between submitting requests to the block io layer to the time when the request is serviced from the disk?

Answer 1

Just checked the code. It measures from the time it issues the request to the device to completion. Depending on the args passed, it starts measuring from blk_account_io_star() function which tracks the requests when it is first queued in the kernel or the default option is to measure from blk_start_request() which tracks when the disk I/O is issued.