HAMMER Update 16-June-2008 - HEADS UP: MEDIA CHANGED AGAIN (2 of 4)

Matthew Dillon dillon at apollo.backplane.com
Tue Jun 17 01:58:24 PDT 2008


    NOTE!  This week (June 16-20 2008) there are going to be several commits
    each of which change the on-media structure and will require recompilation
    of the HAMMER utilities, kernel, and for test filesystems to be
    newfs_hammer'd.

    This is the second of possibly three or four media changes.  I hope
    to get them all done this week, and then that will be it for media
    changes before the 2.0 release.

    Please note that this week testers should expect some instability due
    to the complexity of the media changes, but I don't expect it to last
    beyond the end of the week.

				    KNOWN BUGS

    * There is currently one known failure case that the FSX test finds, which
      I hope to address this week.  It is related to the BMAP code.

    * In-memory record allocations can still blow out system memory and 
      cause a kmalloc panic.

    * bawrite() sometimes panics due to a race (should get this fixed
      tuesday).

			       REMAINDER OF THIS WEEK

    For the rest of this week I will be working on bugs and the remaining
    media changes that I wanted to get in.  In particular going to a larger
    block size (and probably implementing clustered writes), and tuning
    the low level storage manager.

    I believe that going to the larger blocksize will significantly improve
    performance as 64K and 128K single-record writes will cut the B-Tree
    overhead down by 200-800% verses the 16K single-record writes HAMMER
    does now.  This should also, coupled with some work on the low level
    allocator, greatly improve random read performance in the blogbench test.

					    -Matt
					    Matthew Dillon 
					    <dillon at backplane.com>





More information about the Kernel mailing list