pop

pop

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-snprint

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-snprint

rstream: Use snprintf in place of sprintf

Avoid possible buffer overrun.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use snprintf in place of sprintf

Avoid possible buffer overrun.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

new

new

rstream: Use snprintf in place of sprintf

Avoid possible buffer overrun.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

pop

pop

pop

pop

pop

pop

pop

pop

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rstream-opt

pop

pop

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Use separate connections for latency/bw tests

Optimize each connection for either latency or bandwidth
results.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rstream: Add option to specify size of send/recv buffers

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

refresh

refresh

refresh (create temporary patch)

refresh (create temporary patch)

Refresh of rs-def-qpsize

rsockets: Change the default QP size from 512 to 384

Simple latency/bandwidth tests using rstream showed minimal
difference in performance between using a QP sized to 384
entries versus 512.  Reduce the overhead of a default rsocket
by using 384 entries.  A user can request a larger size by
calling rsetsockopt.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

rsockets: Change the default QP size from 512 to 384

Simple bandwidth tests using rstream showed no difference in
performance between using a QP sized to 384 entries versus 512.
Reduce the overhead of a default rsocket by using 384 entries.
A user can request a larger size by calling rsetsockopt.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

rsockets: Simplify state checks

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

rsocket preload: Use environment variable to set QP size

Allow the user to specify the size of the send/receive queues
and inline data size through environment variables:
RS_SQ_SIZE, RS_RQ_SIZE, and RS_INLINE.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

rsocket: Add option to specify size of inline data

Allow the user to override the default inline data size.
We still require a minimum size in order to transfer receive
buffer update message.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

rsockets: Allow user to specify the QP sizes

Add setsockopt options that allow the user to specify the desired
size of the underlying QP.  The provided sizes are used as the
maximum size when creating the QP.  The actual sizes of the QP
are the smaller of the user provided maximum and the maximum
sizes supported by the underlying hardware.

A user may retrieve the actual sizes of the QP through the
getsockopt call.

The send and receive queue sizes are specified separately.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

rsockets: Define options specific to rsockets

Allow a user to control some of the RDMA related attributes
of an rsocket through setsockopt/getsockopt.  A user specifies
that the rsocket should be modified through SOL_RDMA level.

This patch provides the initial framework.  Subsequent patches
will add the configurable parameters.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>

pop

pop

rsockets: Reduce QP size if larger than hardware maximums

When porting rsockets to iwarp, it was discovered that the default
QP size (512) was larger than that supported by the hardware.
Decrease the size of the QP if the default size is larger than
the maximum supported by the hardware.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>