An ugly workaround was removed in 3c33696265 and 159cefb14d. Besides making the code a little cleaner, it was expected that this would improve latency. However, it did not.
Cause for this was a bug that disabled inline messages for all messages. This is fixed in this commit. Now, the anticipated latency improvement can be seen. Compared to 159cefb14d, the latency for samples with 16 values or less is up to 33% better.
Furthermore, the check for signaling is fixed and some typos are fixed.
Before, it waited until it is really connected. That caused problems,
because the send side will start immediately sending if it is connected.
Especially at high rates (>100k) this was a problem.
As discussed in issue #186 and on IM. This function check every 2048th
cycle if the thread should be canceled.
This also removed the need for 'kill -9' in the integration test.
Before this commit, the Infiniband node recreated the address handle for
the remote node during every cycle. Now, it only creates it directly
after it got ah_attr.
* Meta data was not included in the calculation which determines if
sample should be send inline. This caused errors
* Meta data was not substracted from sample->length on receive side
ib_write() and ib_read() now point to the sequence, ts_origin, and format
members of struct sample in a separate scatter/gather element each.
ib_read() measures the time with time_now() (from villas/timing.h) and
sets all flags at receive side.
Closes#152. As described in #182, we will not rearrange the Queue Pairs
for connected mode. As soon as we test many-to-one connections for the
unrealiable connection, we will look again at this issue.
Prior to this commit, we called rdma_disconnect() and waited for a fixed
amount of time. This check was kind of arbitrary. Now, we keep polling
the receive Completion Queue until ib->conn.available_recv_wrs is zero
and all receive samples are thus given back to the framework.
The QP type is dependend on the port space of the RDMA CM ID. If the
RDMA CM ID is set to TCP, the QP has to be set to a RC. If it is set to
UDP, it has to be set to UD.
Node is now able to send data in RDMA_PS_UDP mode. Right now it creates
a new rdma_cm_id for every connection request. We could/should do this
differently
