FlatBuffers与protobuf性能比较

FlatBuffers发布时，顺便也公布了它的性能数据，具体数据请见 Benchmark 。

它的测试用例由以下数据构成" a set of about 10 objects containing an array, 4 strings, and a large variety of int/float scalar values of all sizes, meant to be representative of game data, e.g. a scene format."

   我感觉这样测试如同儿戏，便自己设计了一个测试用例，主要关注CPU计算时间和内存空间占用两个指标，参考对象是protobuf。

   测试用例为：序列化一个通讯录personal_info_list（table），通讯录可以认为是有每个人的信息(personal_info)的集合。每个人信息personal_info(table)有：个人id(uint)、名字(string)、年龄(byte)、性别(enum, byte)和电话号码(ulong)。本来我想用struct表示personal_info(table)，但是struct不允许有数组或string成员，无奈我用table描述它了。相应的idl文件如下：

  //// FILE : tellist.fbs   //// LICENCE : //// MOD :   namespace   as.tellist;  enum   GENDER_TYPE : byte { MALE = 0, FEMALE = 1, OTHER = 2 }  table personal_info { id : uint; name : string; age : byte; gender : GENDER_TYPE; phone_num : ulong; }  table personal_info_list { info : [personal_info]; }  root_type personal_info_list;

   因为要以protobuf做性能参考，列出protobuf的idl文件如下：    

  //// FILE : tellist.proto   //// LICENCE : //// MOD :   package as.tellist;  enum   gender_type { MALE = 0; FEMALE = 1; OTHER = 2; }  message personal_info { optional uint32 id = 1; optional string name = 2; optional uint32 age = 3; optional gender_type gender = 4; optional uint64 phone_num = 5; }  message personal_info_list { repeated personal_info info = 1; }

在内存中构造37个personal_info对象，并序列化之，重复这个过程100万次。

    测试结果如下：

 测试环境：12Core Intel(R) Xeon(R) CPU E5-2620 0 @ 2.00GHz free total used free shared buffers cached Mem: 66081944 65831100 250844 0 182240 46903452 -/+ buffers/cache: 18745408 47336536 Swap: 975864 724648 251216  protobuf三次测试结果： ./bin/tellist_test loop = 1000000, time diff = 14283ms buf size:841  bin/tellist_test loop = 1000000, time diff = 14096ms buf size:841  bin/tellist_test loop = 1000000, time diff = 14229ms buf size:841 占用内存空间841Byte，平均运算时间42608ms / 3 = 14202.7ms  flatbuffers三次测试结果： bin/tellist_test loop = 1000000, time diff = 11694ms buf size:1712  bin/tellist_test loop = 1000000, time diff = 11710ms buf size:1712  bin/tellist_test loop = 1000000, time diff = 11774ms buf size:1712 占用内存空间1712Byte，平均运算时间35178ms / 3 = 11726ms   测试环境：1 Core Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz MEM total used free shared buffers cached Mem: 753932 432672 321260 0 89236 258052 -/+ buffers/cache: 85384 668548 Swap: 1324028 0 1324028  protobuf三次测试结果： bin/tellist_test loop = 1000000, time diff = 12779ms buf size:841  bin/tellist_test loop = 1000000, time diff = 13475ms buf size:841  bin/tellist_test loop = 1000000, time diff = 12604ms buf size:841 占用内存空间841Byte，平均运算时间38858ms / 3 = 12952.7ms  flatbuffers三次测试结果： bin/tellist_test loop = 1000000, time diff = 9424ms buf size:1712  bin/tellist_test loop = 1000000, time diff = 9277ms buf size:1712  bin/tellist_test loop = 1000000, time diff = 9265ms buf size:1712 info vecotor size:37, its right size:37 占用内存空间1712Byte，平均运算时间28036ms / 3 = 9345ms

从以上数据看出，在内存空间占用这个指标上，FlatBuffers占用的内存空间比protobuf多了两倍，而二者的cpu计算时间虽然相差3000ms左右，但考虑到测试用了100万次，二者每次计算时间几乎没有差别。

从以上测试数据来看，FlatBuffers的性能并不如它吹嘘的那么好。个人稍有点失望。