| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
A little benchmarking showed that we were encountering far too many
wakeups, leading to severe performance degradation; we had a bunch
of threads all sleeping on the same condition variable (taskqs)
and this woke them all up, resulting in heavy mutex contention.
Since we only need one of the threads to wake, and we don't care which
one, let's just wake only one. This reduced RTT latency from about
240 us down to about 30 s. (1/8 of the former cost.)
There's still a bunch of tuning to do; performance remains worse than
we would like.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
fixes #38 Make protocols "pluggable", or at least optional
This is a breaking change, as we've done away with the central
registered list of protocols, and instead demand the user call
nng_xxx_open() where xxx is a protocol name. (We did keep a
table around in the compat framework though.)
There is a nice way for protocols to plug in via
an nni_proto_open(), where they can use a generic constructor
that they use to build a protocol specific constructor (passing
their ops vector in.)
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This does a few things. First it closes some preexisting leaks.
Second it tightens the overall close logic so that we automatically
discard idhash resources (while keeping numeric values for next id
etc. around) when the last socket is closed. This then eliminates
the need for applications to ever explicitly terminate resources.
It turns out platform-specific resources established at nni_init()
time might still be leaked, but it's also the case that we now no
longer dynamically allocate anything at platform initialization time.
(This presumes that the platform doesn't do so under the hood when
creating critical sections or mutexes for example.)
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
The use of a single function to get both size and length actually
turned out to be awkward to use; better to have separate functions
to get each. While here, disable some of the initialization/fork
checks, because it turns out they aren't needed.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
The throughput performance tests "try" to avoid hitting the allocator,
but I think this actually causes other cache related performance, and the
receive thread still has to perform a message allocation, leading to really
rotten performance. Its probably time to think about a message pool.
|
| |
|
