We are developing a virtual desktop / remote streaming service which based on Vulkan and OpenGL ES, running on Quadro and Tesla cards.
In our environment, there will be about 20 container instances running on each Linux physical server, and each container instance has 6 - 8 Vulkan processes running simultaneously.
However, it looks like there is a system wide limit of how many Vulkan logical devices can be created, also when the limit is reached, some OpenGL ES functions(like functions from extension GL_EXT_memory_object) just return GL_OUT_OF_MEMORY.
I have tested on Quadro P2000 and Tesla P100 cards, the result shows the limit is around 42 devices, so we can only run about 6 instances simultaneously.
My question is there any way to increase the limit?