On 2020-06-17 12:26, Reco wrote:
On Wed, Jun 17, 2020 at 12:10:51PM -0700, David Christensen wrote:
2. AIUI dd(1) uses asynchronous (buffered) I/O unless told otherwise.
You seem to confuse asynchronous and cached I/O too.
From Linux kernel POV, *asynchronous* I/O is a pair of
io_submit/io_getevents syscalls, and dd does not do these regardless of
the options that are provided.
What dd does is a *synchronous* I/O (read/write syscalls or
pread64/pwrite64 for older kernels).
Whenever the I/O is cached (an output file is opened without
O_DIRECT|O_DSYNC flags) or not is orthogonal to whenever it's sync or
async.
It would not surprise me if there are a dozen layers of chip registers,
caches, buffers, memory, etc., between a CPU register and whatever
device implements data storage in a recent to modern x86_64 computer.
And, I can only wonder if reading /dev/zero even comes from a CPU
register -- it would not surprise me if an MMU feature does.
I was referring to the 'fdatasync', 'fsync', 'dsync', 'sync', and
'nocache' options to dd(1). Given the terse manual page, and a
unwillingness to crawl the dd(1) and/or kernel code, I can only guess at
my understanding of what is happening. But, all of those options seem
to imply some opposite of asynchronous or buffered I/O as I know it.
"The devil is in the details."
For the true async I/O you'll need something like fio, not dd.
That looks like an interesting tool.
David
