Hash-Based Random-Integer find Find Timing Test

Description

This test inserts a number of values with uniform i.i.d. integer keys into a container, then performs a series of finds using find. It measures the average time forfind as a function of the number of values inserted.

(The test was executed with random_int_find_timing_test 200 200 2100)

Purpose

The test checks the effect of different underlying hash-tables (see Design::Associative Containers::Associative Containers::Hash-Based Containers), range-hashing functions, and trigger policies (see Design::Associative Containers::Hash-Based Containers::Hash Policies and Design::Associative Containers::Hash-Based Containers::Resize Policies).

Results

Figures NCCG, NCCM, and NCCL show the results for the native and collision-chaining types in g++, MSVC++, and local, respectively; Figures NGPG, NGPM, and NGPL show the results for the native and probing types in g++, MSVC++, and local respectively.

no image
NCCG: Native and collision-chaining hash random int find timing test using find - g++

In the above figure, the names in the legends have the following meaning:

  1. n_hash_map_ncah- std::tr1::unordered_map with cache_hash_code = false
  2. cc_hash_mod_prime_nea_lc_1div8_1div1_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mod_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_prime_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/1
  3. cc_hash_mod_prime_nea_lc_1div8_1div2_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mod_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_prime_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2
  4. cc_hash_mask_exp_nea_lc_1div8_1div1_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mask_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_exponential_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/1
  5. cc_hash_mask_exp_nea_lc_1div8_1div2_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mask_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_exponential_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2
no image
NCCM: Native and collision-chaining hash random int find timing test using find - msvc++

In the above figure, the names in the legends have the following meaning:

  1. cc_hash_mod_prime_nea_lc_1div8_1div1_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mod_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_prime_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/1
  2. cc_hash_mod_prime_nea_lc_1div8_1div2_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mod_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_prime_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2
  3. n_hash_map_ncah- stdext::hash_map
  4. cc_hash_mask_exp_nea_lc_1div8_1div1_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mask_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_exponential_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/1
  5. cc_hash_mask_exp_nea_lc_1div8_1div2_nsth_map- cc_hash_table with Comb_Hash_Fn = direct_mask_range_hashing , and Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_exponential_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2
no image
NCCL: Native and collision-chaining hash random int find timing test using find - local
no image
NGPG: Native and collision-chaining hash random int find timing test using find - g++

In the above figure, the names in the legends have the following meaning:

  1. gp_hash_mod_quadp_prime_nea_lc_1div8_1div2_nsth_map- gp_hash_table with Comb_Hash_Fn = direct_mod_range_hashing , Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_prime_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2, and Probe_Fn = quadratic_probe_fn
  2. n_hash_map_ncah- std::tr1::unordered_map with cache_hash_code = false
  3. gp_hash_mask_linp_exp_nea_lc_1div8_1div2_nsth_map- gp_hash_table with Comb_Hash_Fn = direct_mask_range_hashing , Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_exponential_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2, and Probe_Fn = linear_probe_fn
no image
NGPM: Native and collision-chaining hash random int find timing test using find - msvc++

In the above figure, the names in the legends have the following meaning:

  1. gp_hash_mod_quadp_prime_nea_lc_1div8_1div2_nsth_map- gp_hash_table with Comb_Hash_Fn = direct_mod_range_hashing , Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_prime_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2, and Probe_Fn = quadratic_probe_fn
  2. n_hash_map_ncah- stdext::hash_map
  3. gp_hash_mask_linp_exp_nea_lc_1div8_1div2_nsth_map- gp_hash_table with Comb_Hash_Fn = direct_mask_range_hashing , Resize_Policy = hash_standard_resize_policy with Size_Policy = hash_exponential_size_policy , and Trigger_Policy = hash_load_check_resize_trigger with αmin = 1/8 and αmax = 1/2, and Probe_Fn = linear_probe_fn
no image
NGPL: Native and collision-chaining hash random int find timing test using find - local

Observations

In this setting, the choice of underlying hash-table (see Design::Associative Containers::Hash-Based Containers ) affects performance most, then the range-hashing scheme (See Design::Associative Containers::Hash-Based Containers::Hash Policies ), and, only finally, other policies.

When comparing Figures NCCG and NCCM to NGPG and NGPM , respectively, it is apparent that the probing containers are less efficient than the collision-chaining containers (both std::tr1::unordered_map and stdext::hash_map use collision-chaining) in this case.

( Hash-Based Random-Integer Subscript Insert Timing Test shows a different case, where the situation is reversed; Observations::Hash-Based Container Types discusses some further considerations.)

Within each type of hash-table, the range-hashing scheme affects performance more than other policies; Hash-Based Text find Find Timing Test::Observations discusses this. In Figures NCCG , NCCM , NGPG , and NGPM , it should be noted that std::tr1::unordered_map and stdext::hash_map are hard-wired currently to mod-based and mask-based schemes, respectively.

Observations::Hash-Based Container Types summarizes some observations on hash-based containers; Observations::Hash-Based Containers' Policies summarizes some observations on hash-based containers' policies.