https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=277992
--- Comment #10 from [email protected] --- Created attachment 250268 --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=250268&action=edit write performance of the tests starting at 16:00 on the graph, I sent a series of zfs send | zfs recv of the zfs data set and zfs destroy the copy. I pause for 120 seconds, then initiate a manual trim zpool trim -w pool and then pause 120secs and then start again. As you can see the throughput is consistent on writes. Then at 08:00 I restart the test, but this time I dont do a trim after the zfs destroy. If manages about 3 times (the dataset is about a 1/3rd of the pool's capacity and then the writes become brutally slow which tracks FreeBSD. If I then do a zpool trim -w pool, performance is back to normal. -- You are receiving this mail because: You are the assignee for the bug.
