On 2019/11/13 2:10 下午, Christian Balzer wrote: > On Wed, 13 Nov 2019 11:44:50 +0800 Coly Li wrote: > > [snip] >> >> Hi Christian, >> >> Could you please try the attached patch in your environment ? Let's see >> whether it makes things better on your side. >> > > Don't have custom/handrolled kernels on those machines, but I'll give it a > spin later. > Looking at the code I'm sure it will work, as in not going to full speed > when idle. > > Is there any reason for this being a flag instead of actually setting the > max writeback rate, as mentioned when comparing this to MD RAID min/max? > > What this does now is having writeback_rate_minimum both as the min and max > rate for non-dirty pressure flushing. > Whereas most people who want to actually set these values would probably > be interested in a min rate as it is now (to drain things effectively w/o > going overboard) and a max rate that never should be exceeded even if the > PDC thinks otherwise. For the max writeback rate limit, so far it is handled by the PDC controller. I will have a try whether I can make it myself before anybody helps to post patch. But it is at quite low priority location in my todo list, I need to complete other tasks firstly which are not easy neither and spent a lot of time already. I post the patch to upstream for Linux v5.5, and add Reported-by: tag with your email address. Thanks. -- Coly Li
