Hi all, For anyone who interests, I've made a preliminary kernel implementation and it seems to work now. I'm cleaning up these patchset and hopefully send it out later. The code can be fetched from:
git://git.kernel.org/pub/scm/linux/kernel/git/xiang/erofs-utils.git -b experimental-bigpcluster and git://git.kernel.org/pub/scm/linux/kernel/git/xiang/linux.git -b erofs/bigpcluster The benefit has been showed in the previous email, see: https://lkml.kernel.org/r/20201230101239.ga3282...@xiangao.remote.csb What I currently saw is that the sequential read can have slightly benefit compared with EROFS 4k case (mainly due to CR improvement), correspondingly rand9m has some drop, that is understandable. The next step would be to enable compact indexes for big pclusters. That will boost up the read performance even further. Also there are still some finer optimization for big pclusters as well. I will seek more extra time on optimizing case-by-case. I hope the big pcluster on-disk format can be finalized in the next month, and upstream the first runtime version for the next 5.13 cycle. Finally, first EROFS LZMA compression implementation will be planned for the 5.14 cycle if no strange happens. With bigpcluster feature landed, EROFS LZMA compression can be more powerful then since such end users care more about CR rather than random performance instead. Thanks, Gao Xiang