first I want to say that I really enjoyed all the great presentations and
the conversations at the SPDK Summit! A conference like that is really
I'm currently looking into integrating SPDK (mainly nvme-of) into an
environment that has some additional constraints and I was hoping to get
The first issue is related to spdk_mem_pool which used for the nvmeof
iovecs that are passed to the bdev layer
Is there a way to allow a bdev to allocate these? Or do I need to implement
my own spdk_mem_pool via the environment abstraction? I have a requirement
where the iovec memory needs to be accessible by other, non-SPDK processes
outside of the SPDK process.
The second question is related to the dynamic (spdk_) thread scheduling
that Ben presented. I would like to schedule threads dynamically to a
different core based on how loaded the cores are. In other words, if a core
is super busy, move the (spdk_) thread to another core.
Our threading environment is cooperative and the spdk_thread_poll function
would work great but it is unclear what do to
with _spdk_event_queue_run_batch() that processes reactor events.
This seems to be tied 1:1 to a reactor core. It is also unclear to me how I
would create our own reactor given that spdk_event_allocate/call API is in
Show replies by date