Hitoshi Mitake

@sa-kura thanks a lot for your analysis. Is it possible to evaluate performance after your change?

@vtolstov I'm not sure yet. But if you can benchmark https://github.com/sheepdog/sheepdog/tree/per-cpu-work-queue with rebasing on the master, it will be helpful.

@gangzai sorry for my late reply. Are you trying to backup hypervolume with cluster snapshot?

cc @vtolstov

For example, - workqueue status (a number of worker threads, load, min/max/average of completion time, etc) - sockfd cache status - a number of requests per opcode - etc... and...

@pttpzp We had another caching mechanism (based on cache of disk drive) that is already removed. If you really want a caching mechanism that works correctly, we can revive it....

@tmenjo If you keep the flags for compatibility, removing the features would be transparent and won't make troubles visible from users.

@tmenjo on the second thought you can remove the flags because v1.0 doesn't need to be fully compatible with older versions

@pttpzp of course utilizing SSD is important for storage systems. If you have an idea, please let us know :)

@imrook , can someone@NIMBOXX work on this?