On Thu, Nov 29, 2018 at 05:15:33PM +0800, Coly Li wrote: > On 11/29/18 4:34 PM, Vojtech Pavlik wrote: > > On Thu, Nov 29, 2018 at 03:35:07PM +0800, Coly Li wrote: > >> On 11/29/18 11:16 AM, fangchen sun wrote: > >>> hello guys, > >>> > >>> I want to change writeback_percent in bcache, but it seems that the > >>> value is hardcoded in the bcache drive code. Why is writeback_percent > >>> capped at 40 ？ Can I set it 60 or higher percent? What are risks of > >>> higher percent ? > >> > >> I just posted a 5 patches series titled "writeback performance tuning > >> options", which may permit people to set writeback percent to 70 at most > >> for research purpose. > >> > >> Generally I don't suggest to set a higher writeback percent, a lower > >> writeback percent number may start writeback thread earlier and have a > >> more smooth I/O latency number. Too much dirty data on cache device may > >> have negative impact for most of work load. > >> > >> You may try the patch set with a higher writeback percent, but unless > >> you have a strong reason, I will always to use the default values firstly. > > > > One thing I'm wondering about: The mode (writeback/writethrough) of the > > device is persistent, while the writable percentage and other tunables > > need to be reset every boot (registration). It'd be very nice if all the > > parameters of the bcache device were persistent. > > > > Yes, this is one thing on my ToDo list, the rested items in the list are, > - journal hang/locking issue > - s390x support > - 4Kn support > - partition enhancement > - bigger super block (this is for persistent parameters) > - multiple internal B+ trees > - remove multiple cache devices code framework > - generate entries in /dev/disk/by-**/ directories Ah, I see. That's a good list of future features. -- Vojtech Pavlik Director SUSE Labs
