Components

Functions

The initial implementation (basic.py)

We started with a straightforward single-threaded Python script:

• File I/O:

  Used plain open() line by line with UTF-8 decoding.

• Parsing:

  Used float() for each temperature measurement, multiplied by 10 for fixed-point.

• Aggregation:

  A standard Python dictionary with get() and __setitem__ to maintain [count, sum, min, max].

• Garbage collection disabled to avoid pauses.

Performance:

~10.5 seconds on 10 million lines.

This was expected: most time spent was in

• repeatedly allocating strings,
• parsing floats in Python (slow!),
• updating dictionary aggregates.

Step 1: Memory-mapped I/O (prime.py)

We tackled the biggest bottleneck first — file reading.

What we changed

• Used Python’s mmap to memory-map the entire file.

  Now file contents live in virtual memory, and no Python strings are created per line. The OS handles paging, and seeking or slicing costs almost nothing.