kangle 3.6.0版本新增boost fcontext协程，应该能比之前的ucontext进一步提高多线程切换性能
kangle 3.6.0版本新增boost fcontext协程，应该能比之前的ucontext进一步提高多线程切换性能表现在多核心CPU上运行的kangle，使用fcontext应该能提高性能
https://www.boost.org/
https://github.com/boostorg/context
原用的ucontext
https://github.com/kaniini/libucontext
编译方法
cmake .. -DENABLE_FCONTEXT=1
新的fcontext测试对比多线程性能大约提升11%
ucontext
Concurrency Level: 100
Time taken for tests: 10.829 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Total transferred: 48314024 bytes
HTML transferred: 23500000 bytes
Requests per second: 9234.44 [#/sec] (mean)
Time per request: 10.829 (mean)
Time per request: 0.108 (mean, across all concurrent requests)
Transfer rate: 4356.96 received
fcontext
Concurrency Level: 100
Time taken for tests: 10.063 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Total transferred: 48313336 bytes
HTML transferred: 23500000 bytes
Requests per second: 9937.07 [#/sec] (mean)
Time per request: 10.063 (mean)
Time per request: 0.101 (mean, across all concurrent requests)
Transfer rate: 4688.41 received
3.5.21.16生产版本，看来拿3.6.0开发版本对比生产版本果然是，，，毕竟生产版本有jemalloc高性能加持
Concurrency Level: 100
Time taken for tests: 7.525 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Total transferred: 48381928 bytes
HTML transferred: 23500000 bytes
Requests per second: 13288.51 [#/sec] (mean)
Time per request: 7.525 (mean)
Time per request: 0.075 (mean, across all concurrent requests)
Transfer rate: 6278.55 received
两个3.6开发版对比，，确实强了一丢丢，感觉是误差，测了几次结果大概5-7%左右提升
感觉可能是编译没有启用-DCMAKE_BUILD_TYPE=Release选项导致的，启用后在测了一次，，，发现结果依旧一样没区别，性能比3.5.21.16下降了30%左右性能
3.5.21.16
Concurrency Level: 100
Time taken for tests: 7.234 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Total transferred: 24644184 bytes
HTML transferred: 0 bytes
Requests per second: 13823.46 [#/sec] (mean)
Time per request: 7.234 (mean)
Time per request: 0.072 (mean, across all concurrent requests)
Transfer rate: 3326.83 received
ucontext
Concurrency Level: 100
Time taken for tests: 9.313 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Total transferred: 24616784 bytes
HTML transferred: 0 bytes
Requests per second: 10738.10 [#/sec] (mean)
Time per request: 9.313 (mean)
Time per request: 0.093 (mean, across all concurrent requests)
Transfer rate: 2581.42 received
fcontext
Concurrency Level: 100
Time taken for tests: 9.349 seconds
Complete requests: 100000
Failed requests: 0
Write errors: 0
Total transferred: 24604296 bytes
HTML transferred: 0 bytes
Requests per second: 10696.76 [#/sec] (mean)
Time per request: 9.349 (mean)
Time per request: 0.093 (mean, across all concurrent requests)
Transfer rate: 2570.18 received
2024年8月21日补充
官方已经优化修复3.6.0性能下降的问题，请更新到最新提交的代码，编译打开选项 -DENABLE_JEMALLOC=1 也可以带来10%的性能提升
https://github.com/keengo99/kangle/commit/d7f955d5b038e72e94b17b001a85f9f4ebf2a56f
但是经过我测试依旧不行，，，性能对比3.5来说还是很差
3.6新版是重构为动态模块化的 编译好的文件不能直接用 要在机器上单独编译使用
我这编译好的二进制文件发你 你那是无法执行的 会提示so文件未找到
这也可能是性能下降的一个原因，以前是静态编译文件50M，现在动态的只有2M
2024年11月29日补充
今天3.6定版发布更新后，优化了网络发送和接收，减少了上下文切换，性能做到和3.5基本一致了，但是我感觉还有一些优化空间，估计要启用iouring模型
版本 3.5.21.16(enterprise)，最新版的kangle去哪里装哦，我现在都用的一键脚本 maxrate 发表于 2024/8/11 13:00
版本 3.5.21.16(enterprise)，最新版的kangle去哪里装哦，我现在都用的一键脚本
3.6.0目前处于开发阶段，感兴趣可以去官方仓库下载源代码来使用
https://github.com/keengo99/kangle
页:
[1]