Wayne Gao

> The issue is that you are on much newer compiler and os kernel than what bam supports. > > We don't intend to update the codebase to support newer...

> Many have successfully reproduced results following exact steps and version described in the readme. I encourage to try that. thank you so much man. now I tried ubuntu 20.04.3,...

 when I call io_uring_peek_cqe, it will invoke syscall everytime.

@leonrayang @xiaochunhe hello buddy, I write one C++ Fuse mock tool that can be used to experiment different fuse mode and parameters. this mock fuse file system, read will translate...

@leonrayang @xiaochunhe could you please some guide and help that how to contribute? :)

> [@gaowayne](https://github.com/gaowayne) Thanks for your work, does the fio result show the pure memory based on the fuse of CubeFS? If you want to contribute to CubeFS, please write a...

@leonrayang @Victor1319 guys, is there any idea I can tune the performance. is this fuse overhead on our code? now I increase the jobs number, but ingestion still only 400MB/s...

@leonrayang my NVMe can offer 14GB/s read BW, is there any way we can tune parameter to use higher BW for fio read workload? ``` test_file_c: Laying out IO files...

> -direct=1 是没有缓存，直接从datanode上去读取的。这个性能都比较低。 那，如果direct=1，这个性能是基本差不多了？其实我的NVMe盘性能非常高，通过page cache，其实反倒会浪费CPU拉低性能。那咱们可以一起调优一下，目前空间非常大。

> 对的。建议你考虑一下bcache，或者上面的那个ahead read 我是random read, ahead read效果应该不会太好，bcache应该对random read也不是cache friently, 我enable page cache的话，就是测内存了，我其实就想摸底存储系统，对高性能介质的情况