Interesting, I wrote a modified version where you can enter the string length and the sentinel version is consistently slower (using gcc-4.4.0 with -O3). // sentinel terminated sequence iterator proof-of-concept // Copyright Beman Dawes, 2009 #include <cassert> #include <cstring> #include <iostream> #include <boost/functional/hash.hpp> #include <boost/timer.hpp> // sentinel terminated sequence iterator template < class T, T Sentinel = T() > class sentinel_iterator { public: sentinel_iterator() : m_ptr(0) {} // construct end iterator sentinel_iterator( T * p ) : m_ptr( (p && *p != m_sentinel) ? p : 0 ) {} sentinel_iterator & operator++() { assert(m_ptr); if ( *++m_ptr == m_sentinel ) m_ptr = 0; return *this; } T & operator*() const { assert(m_ptr); return *m_ptr; } bool operator==( const sentinel_iterator & rhs ) const { return m_ptr == rhs.m_ptr; } bool operator!=( const sentinel_iterator & rhs ) const { return m_ptr != rhs.m_ptr; } private: T * m_ptr; // 0 == the end iterator static const T m_sentinel = Sentinel; }; static size_t test_strlen( const char* s ) { return boost::hash_range( s, s + strlen( s ) ); } static size_t test_sentinel( const char* s ) { return boost::hash_range( sentinel_iterator< const char >( s ), sentinel_iterator< const char >() ); } int main() { const unsigned N = 1000000; std::cout << "enter text length: "; unsigned len; std::cin >> len; char* text = new char[len]; for (unsigned i = 0; i < len; ++i) text[i] = '4'; text[len] = '\0'; std::size_t m1 = 0; std::size_t m2 = 0; { boost::timer t; for( unsigned i = 0; i < N; ++i ) { m1 += test_strlen( text ); } std::cout << "strlen: " << t.elapsed() << " " << m1 << std::endl; } { boost::timer t; for( unsigned i = 0; i < N; ++i ) { m2 += test_sentinel( text ); } std::cout << "sentinel: " << t.elapsed() << " " << m2 << std::endl; } delete[] text; return 0; }