diff --git a/protocol/README.md b/protocol/README.md index 7af5708a..27604cfc 100644 --- a/protocol/README.md +++ b/protocol/README.md @@ -53,13 +53,11 @@ memory：64g - Use Javassist bytecode to enhance the dynamic generation of serialization and deserialization functions for sequential execution, and sequential functions can be easily JIT compiled to achieve extreme performance -- Natively integrated with netty's high-performance Byte Buf +- Natively integrated with netty's high-performance ByteBuf, support Zero Copy - With primitive type collection, there is no boxing and unboxing, invalid GCs are avoided, and the performance is fast enough -- Inherently thread-safe and lock-free; kryo forces each thread to have its own instance of Kryo, which is a heavy - design, especially in scenarios with many threads -- No reflections, no unsafe operation; The use of objenesis in Kryo resulted in a lot of unsafe, and warnings when - running in Java 11 +- thread-safe and lock-free, without any performance loss in multi-threaded environment +- No reflection, no unsafe operations, support GraalVM - Flattening the call depth of the method stack, and there is no performance penalty for nesting data structures, such as List>>; Comparing kryo and protobuf data structure nesting results in a performance penalty - There is no risk of vulnerability injection, only bytecode enhancement will be performed during initialization, and no diff --git a/protocol/README_CN.md b/protocol/README_CN.md index aef8fbcd..50331a65 100644 --- a/protocol/README_CN.md +++ b/protocol/README_CN.md @@ -45,10 +45,10 @@ cpu： i9900k ### Ⅳ. 为什么快 - 使用Javassist字节码增强动态生成顺序执行的序列化和反序列化函数，顺序化的函数可以轻易的被JIT编译以达到极致的性能 -- 原生集成netty的高性能ByteBuf +- 原生集成netty的高性能ByteBuf，支持Zero Copy - 使用primitive type collection，没有装箱和拆箱，避免了无效GC，性能也足够快 -- 天生线程安全并且无锁化；kryo强制要求每条线程都有自己的一个Kryo实例，这是一个比较重的设计，特别是线程比较多的场景 -- 没有反射，没有unsafe操作；对比kryo中使用objenesis导致大量unsafe，而且在Java11中运行会出现警告 +- 天生线程安全并且无锁化，多线程环境无任何性能损失 +- 没有反射，没有unsafe操作，只是GraalVM - 扁平化了方法栈的调用深度，数据结构嵌套没有任何性能损失，如List>>；对比kryo和protobuf数据结构嵌套会出现性能损失 - 无漏洞注入风险，只有初始化时会进行字节码增强，后期不会再进行任何字节码的操作 - 充分利用程序局部性原理，序列化器放在short[]数组中，对象序列化和反序列化只会查找一次序列化器