Why Text Files Are Slow: Faster C++ Banking Ops
When building BankEase, a banking simulation engine, I initially used standard text file I/O — fstream with << operators — to store user records. It was readable, but slow.
As the dataset grew to thousands of accounts, the overhead of parsing text (converting the string "1000.50" into the float 1000.50) became the bottleneck. The disk wasn’t the problem. The parser was.
The solution: binary I/O
I rewrote the storage engine to use binary serialisation. Instead of writing human-readable characters, I dumped the raw memory of the C++ objects directly to disk.
The old way — slow parsing:
file << account.id << " " << account.balance << endl;The new way — raw memory copy:
file.write(reinterpret_cast<char*>(&account), sizeof(Account));This single change achieved 300% faster read/write operations for account transactions.
The trade-offs worth knowing
Binary I/O isn’t free. The file is no longer human-readable, and reinterpret_cast over a whole struct only works while the type stays trivially copyable — no pointers, no std::string members, no virtual functions. It also bakes in your platform’s padding and endianness, so a file written on one machine isn’t guaranteed to load on another.
For a self-contained simulation those constraints are fine, and the speed is worth it. For anything crossing machines, you want a real serialisation format instead.