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。