HAMMER update 05-May-2008

Johannes Hofmann Johannes.Hofmann at gmx.de
Wed May 7 08:48:35 PDT 2008


Hi,

just wanted to check out HAMMER once again in a vkernel. But I kept
getting errors like:

# mount_hammer /dev/vkd1 /dev/vkd2 /mnt/
# cpdup /usr/ /mnt/
/mnt//bin: mkdir failed: Unknown error: 1711798680
/mnt//include/crypto/castsb.h    rename-after-copy failed: No such file or directory
/mnt//include/c++/3.4/bits/stl_deque.h rename-after-copy failed: No such file or directory
/mnt//include/c++/3.4/algorithm  rename-after-copy failed: No such file or directory

almost immediately. Later the vkernel crashed sometimes.

However, this only seems to happen with gcc34 compiled vkernels!
If I compile them with gcc 4.1.2 everything seems fine.
Is this expected?

  Johannes




Matthew Dillon <dillon at apollo.backplane.com> wrote:
>    I am now working on the filesystem full handling.   Everything else is
>    working and appears to be stable.
> 
>    Current issues:
> 
>    * There may be a few undo cases where recovery-after-crash is
>      unsuccessful, or results in an assertion later on.  I found
>      a big issue with the kernel itself trying to sync the buffers
>      and I whacked that (the last thing HAMMER wants is for the kernel
>      to sync its buffers!).  I am continuing to test by intentionally
>      crashing the system both under a vkernel and on real test boxes.
> 
>    * Reading and writing are not optimized well yet.  For reading the issue
>      is a lack of clustering and read-ahead support (yet).  For writing
>      the issue is due to the way buffers are staged out to the backend.
> 
>      Both issues are fairly straight forward, but they aren't high
>      priority at the moment.
> 
>    * The filesystem full handling isn't in yet.  That is next on my list.
> 
>    I have begun testing HAMMER on real hardware.  My backup machine is now
>    running a 700GB HAMMER filesystem.  I expect more issues will work their
>    way out of the woodwork but right now things are looking *very* good.
> 
>                                        -Matt
>                                        Matthew Dillon 
>                                        <dillon at backplane.com>





More information about the Kernel mailing list