hyb

I have a good implemention of spare hash map https://github.com/ktprime/emhash/blob/master/hash_table8.hpp it's the fastest iteration compared with other's hash map, but it's not efficient for erasion/insertion if Key/Value is large. it...

> I tried windows platform + clang compiler. robin_hood performs as fast as MSVC. The performance dropdown of linux seems casued by OS instead of compiler. you can try my...

> I have tried Windows/MSVC vs Apple/Clang on machines with similar processors, and Clang wins always (sometimes a lot). Worth noting: ska hashmaps beat robin_hood almost always. I guess (?)...

有必要优化网络收发包定时器性能，大多情况下定时器不会被触发（重传，ping，idle,）采用惰性删除（标记失效）。下次加入直接从当前位置上下调整，可以大量减少不必要的堆操作。 我在google quic项目中优化了定时器性能，测试100个活跃连接，服务器小堆定时器每秒达30w/s级别弹出和插入，优化后性能大幅度提升，perf top看不到任何堆耗时操作了。

I have updated your old test code and add some new hash map benched with different k-v/cpu/os/compliers/memory combination. The result is quite different (https://github.com/ktprime/emhash/tree/master/bench/tsl_bench) . I also bench them with...

I can reproduce it now in my test(gcc only) and test code https://github.com/ktprime/emhash/blob/master/bench/btest.cpp //download my git emhash and run test with gcc 7.5 root@ubuntu:~/emhash/bench g++ -O3 -march=native -mtune=native -I.. -I../thirdparty...

> 32GB server. it's a bad hash function cause many rehash ?

如果只是毫秒精度下面这个足够快 ``` int64_t rdns() { timespec ts; ::clock_gettime(CLOCK_MONOTONIC_COARSE, &ts); return ts.tv_sec * 1000'000'000 + ts.tv_nsec; } ```

多路上面可能不准，通常情况下要定期校验， 多线程情况下周期性调用init 是否有问题？ include 在macOS clang++上无法编译，似乎已经不是c++标准头文件了（gcc 保留）

服务器跨平台使用修改rdsysns，使用steady_clock 是否比high_resolution_clock 更稳定 还是没太大区别? 改成steady_clock 后，测试数据不一样，b2c 数据误差持续变大 a: 14642789863999881, b: 14642789864006251, c: 14642789864006229, a2b: 6370, b2c: -22, good: 0, rdsysns_latency: 6339 a: 14642790864015103, b: 14642790864021136, c: 14642790864021107, a2b: 6033, b2c:...