Profiling PHP

I did a quick little profile of a lightly hit PHP server running a variety of PHP apps such as IMP, Gallery and a couple of small MySQL-driven apps. A semi-representative tiny snapshot of what I would consider normal usage of PHP. I threw oprofile at it (oprofile.sourceforge.org) and here are the results: Cpu type: PIII Cpu speed was (MHz estimation) : 868.000000 Counter 0 counted CPU_CLK_UNHALTED events (clocks processor is not halted) with a unit mask of 0x00 (Not set) count 600000 vma samples %-age symbol name 001a56bc 1018 19.7096 zendparse 001a9f4c 369 7.14424 lex_scan 001b1ea4 349 6.75702 _emalloc 001d4ca0 326 6.31171 execute 001b21b8 301 5.82769 _efree 001cdaf8 238 4.60794 zend_inline_hash_func 001ca85c 238 4.60794 zend_hash_add_or_update 001f5960 171 3.31075 mail_msgno 001b3028 170 3.29138 _mem_block_check 0015dcb4 127 2.45886 php_var_unserialize 001ccae0 120 2.32333 zend_hash_find 001baf6c 81 1.56825 _zval_ptr_dtor 001c443c 69 1.33591 _zval_dtor 0015701c 66 1.27783 smart_str_print_unsigned 001cbda8 53 1.02614 zend_hash_destroy 001db270 51 0.987415 _get_zval_ptr 001ca5f8 48 0.929332 _zend_is_inconsistent 00156c08 43 0.832527 smart_str_appendl_ex 001cb9b8 41 0.793804 zend_hash_rehash 001cc6cc 39 0.755082 zend_hash_copy 001f64d0 36 0.696999 mail_gc_msg 0015ef14 33 0.638916 process_nested_data 001563b0 33 0.638916 php_var_serialize_intern 001b27fc 31 0.600194 _estrndup 00208200 29 0.561471 mm_blocknotify 0015ee18 28 0.54211 var_push 001c45c0 27 0.522749 zval_add_ref 0015f2b4 26 0.503388 parse_iv2 001bdaf0 25 0.484027 destroy_op_array 001c4970 24 0.464666 _zval_ptr_dtor_wrapper 001cb3b4 22 0.425944 zend_hash_index_update_or_next_insert 001b2440 21 0.406583 _erealloc 001c3560 20 0.387222 zend_str_tolower 001b9970 20 0.387222 zendlex 001c3dfc 19 0.367861 zend_ptr_stack_n_push 001d398c 18 0.3485 zend_fetch_var_address 001c83cc 17 0.329138 add_next_index_long 001c45d0 17 0.329138 _zval_copy_ctor 001bdcd0 16 0.309777 init_op Not all that useful. Would be cool to run this for longer on a busy site to get more accurate numbers. It does give some indication of where small code optimizations would make the biggest overall difference. -Rasmus