Dear colleagues,
Issue (in fork repository) with performance tests and results.
We discussed about that problem with Lourival and decided to try use pool(9).
Firstly, I dived into current kmem_intr_alloc/free implementation that already using pool internally. Analyzing of code shows that current implementation of lua_alloc function that using kmem_intr_alloc/free internally (that use pool internally :)) seems efficient because LUAL_BUFFERSIZE less that maximum allocatable chunk of memory in kmem allocator pool (16KiB).
I have some ideas and I would like to discuss about them.
- kmem allocator implementation seems very efficient and
lua_alloc consequently too. But we have performance problem. It seems that something wrong with my hands. :) But all aims that problem not in I/O operations...
- At that time I creating minimal test code and want to simulate
lua_alloc calls as it was. After that compare results with userland similar tests.
- May be you have some ideas or hypothesis? I spend a lot of time on experiments and measurements. Currently, I have enough time to work on our project. ;)
Dear colleagues,
Issue (in fork repository) with performance tests and results.
We discussed about that problem with Lourival and decided to try use pool(9).
Firstly, I dived into current kmem_intr_alloc/free implementation that already using pool internally. Analyzing of code shows that current implementation of lua_alloc function that using
kmem_intr_alloc/freeinternally (that use pool internally :)) seems efficient becauseLUAL_BUFFERSIZEless that maximum allocatable chunk of memory in kmem allocator pool (16KiB).I have some ideas and I would like to discuss about them.
lua_allocconsequently too. But we have performance problem. It seems that something wrong with my hands. :) But all aims that problem not in I/O operations...lua_alloccalls as it was. After that compare results with userland similar tests.