Interactive performance, memory footprint, and scalability analytics for MolSysMT.
Fetching baseline results...
Median execution time (milliseconds) comparing MolSysMT (Public & JIT), MDTraj, and MDAnalysis on standard operations (Chicken Villin system, 4369 atoms).
MolSysMT's selection query parser resolves complex selections in 8.54 ms vs MDTraj's 49.67 ms.
Dynamic peak RAM delta (MB) allocated during execution, tracking isolated process overhead.
Comparing the execution time of raw compiled Numba math kernels against the public API wrappers (which perform PyUnitWizard wrapping and @arg_digest validation).
Raw JIT Kernels are Extremely Fast
Calculating center coordinates or RMSD on the GPU/JIT runs in ~0.9 ms to 1.1 ms, matching or beating native compiled C++ libraries.Public Wrapper Tax
Resolving dynamic physical units and decorator arguments introduces a timing tax of **~250 ms to 280 ms** on standard wrappers.Developer Best Practice
Internals, pipelines, and sister libraries (like TopoMT) bypass the public wrapper layer and query raw mathematical JIT submodules directly.Average public-to-kernel timing ratio
Processing times (seconds) comparing Eager Loading, Increment-Streaming Iterators, and the ChunkedExecutor wrapper on a solvated protein trajectory (4369 atoms).
Eager Loading (Fastest, ~164 ms)
Best for small/medium files. Issues a single disk read pass and maps memory directly, avoiding loop boundaries.Iterator (Balanced, ~225 ms)
Streams coordinates frame-by-frame. Uses ~37% more time than eager due to repeated I/O bounds, but caps peak RAM.ChunkedExecutor (Heavy, ~356 ms)
Designed for massive out-of-core trajectories exceeding RAM. Includes slicing and Reducer boundaries.Microsecond timings for mechanical vs topological attribute checks, comparing active validation vs. skip digestion (Fast-Path).
Comparative microsecond latency of parsing, dimensionality checking, and extracting values (with vs. without conversion).