VmData usage is 65 MiB per IO thread

[Arun Chandrasekaran]

I did some analysis and I found that the process VmData size is directly proportional to the number of IO threads.

VmData usage is exactly equal to 67,108,864 bytes per IO thread.

I've captured the details here in https://github.com/cpp-netlib/cpp-netlib/issues/771

Any feedback is appreciated. Is there a way to reduce this 65 MB to probably 10 MB or so?

I would like to have a large pool of threads and this is definitely a concern.

Cheers,
Arun

[Dean Michael Berris]

On 1 Aug 2017, at 03:19, Arun Chandrasekaran <vision...@gmail.com> wrote:

I did some analysis and I found that the process VmData size is directly proportional to the number of IO threads.

VmData usage is exactly equal to 67,108,864 bytes per IO thread.

I've captured the details here in https://github.com/cpp-netlib/cpp-netlib/issues/771

Any feedback is appreciated. Is there a way to reduce this 65 MB to probably 10 MB or so?

That's an interesting question. I would think looking into Boost.ASIO's configuration options on how large the queues are for io_service instances might show something. I currently don't have time to spare looking into this myself, but it might be worth exploring if you do.

I would like to have a large pool of threads and this is definitely a concern.

Why do you want a large pool of threads?

Also, why do you need to reduce virtual memory address usage?

-- Dean