Hi there,
When I took a look at the source codes, it shows the ctx_t class implementation uses the pthread mutex locking mechanism. In general, it works fine in many applications. For the SMP architecture, the system runs the multicore and each worker thread can acquire the CPU to run for the completeness at the same time. If use the Zeromq for the "inproc" communication, the inter-thread transport doesn't need the I/O thread, and all messages are in the memory. It's nice to have the spin_lock implemented inside the Zeromq, instead of pthread mutex for this situation. It can reduce the context switch to cause the performance penality. Do you have any idea to work around or plan to support it? Thanks. -Richard
_______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
