c# - Fast parallel adding to dictionary when keys never collide -

i have situation, dictionary created, key-value pairs added, , after point dictionary used reading values.

i'm trying implement quickest way of adding dictionary in initialization phase.

concurrentdictionary has slow tryadd method (and getoradd) - on 6 core cpu (12 threads), cpu usage stays @ 25%, indicating 3 threads used.

it faster add keys (around 25 million) dictionary sequentially, using concurrentdictionary , parallel.for.

how can improve speed here? easy separate key-value pairs, keys never collide when added dictionary, using code below creates problems:

dictionary<long,string> d = new dictionary<long,string>(); d[key] = value; 

it seems this, when used in multi-threaded environment, fails because dictionary internal changes (resizing?).

would work instead?

capacity = 250000000 //basically big enough store data dictionary<long,string> d = new dictionary<long,string>(capacity); d[key] = value; 

i prefer using dictionary on concurrentdictionary, because reads faster (and speed crucial application).

if know maximum size of dictionary, yes can improve speed pre-allocating. however, won't prevent concurrent adds potentially messing dictionary. internal lists updated, , it's quite possible 2 threads end storing key @ same index in list.

have tried obvious thing? is:

lock (dictionarylock) {     dict[key] = value; } 

if lock not contended, it's going take perhaps 20 nanoseconds. if is contended, delay longer, you're doing short operation. whether faster depend lot on kind of processing done between calls add.