[Bug 197516] TRIM on gmirror is slow and results in inresponsive system

bugzilla-noreply at freebsd.org bugzilla-noreply at freebsd.org
Wed Feb 11 10:26:32 UTC 2015


--- Comment #2 from Robert Schulze <rs at bytecamp.net> ---
I've added a debug statement to sys/ufs/ffs/ffs_alloc.c to compare the calls
between a non-mirrored ufs and a gmirrored one:

--- sys/ufs/ffs/ffs_alloc.c.orig        2015-02-09 16:24:06.000000000 +0100
+++ sys/ufs/ffs/ffs_alloc.c     2015-02-11 10:17:13.000000000 +0100
@@ -2313,6 +2313,9 @@
        bip->bio_length = size;
        bip->bio_caller2 = tp;

+       printf("%s: starting BIO_DELETE for %jd bytes @ offset %jd\n", 
+           __func__, size, bip->bio_offset);
        mp = UFSTOVFS(ump);
        vn_start_secondary_write(NULL, &mp, 0);
        g_io_request(bip, (struct g_consumer *)devvp->v_bufobj.bo_private);

Creating a 100m file and deleting it afterwards results in the same amount of
calls to BIO_DELETE in both cases.

Changing kern.geom.dev.delete_max_sectors does not influence the calls.

You are receiving this mail because:
You are the assignee for the bug.

More information about the freebsd-bugs mailing list