perf: reuse backend and allocator between different sites

[lkdvos]

Currently, we are using a buffer allocator by default, which is being reallocated between all sites.
Especially for finite systems this is unnecessary, and we should simply reuse the same buffer constantly.