[OpenAFS] any hack to get multiple read/write mirrors
Martin Schulz
schulz@iwrmm.math.uni-karlsruhe.de
23 Oct 2001 09:45:34 +0200
Zachary Denison <zacharydenison@yahoo.com> writes:
> This is not true. It is my understanding, that in the
> case of of an outage, each office can continue to work
> with "as much of the dataset" that they have, and they
> can continue to make changes to it, and then when the
> network connectivity comes back, that machine which
> was out, will repropagate its changes.
If all workers in all offices will need to access all pieces of data
randomly, then AFS may not be the right choice for you.
AFS is based on rather pessimistic assumptions (it always considers
the worst case) about data cohenrency. Therefore, if an action is not
guaranteed to be fulfilled correctly, then it will not be accepted.
If you write to a file and close it, then you can be sure that your
changes (to the RW volume) are visible anywhere. If AFS cannot ensure
this (maybe because of any network outage between client and server),
it wont let you write the file. The changes will be visible in the RO
volumes only after the appropriate "vos release" command.
There are other filesystems such as CODA or Intermezzo that are far
more optimistic at that point, but require user interaction in the
worst cases of coherency violation (i.e. different users changing the
same file without knowing from each other). From hearsay, I was told
that these are not recommendable for production systems at this time,
YMMV.
> No, because r/o data is useless to the people. At
> least of I server r/w data during an outage people can
> still work with data they have, they just wont see any
> "new data" which is fine.
"the data they have" would be (for the AFS case) the local RW volumes
and the RO replicas of the other offices. The location of the RW
volumes can be changed upon request. If that does not suffice, you
probably need something else than AFS. Don't know much about DCE/DFS
and other developments.
Yours,
--
Martin Schulz schulz@iwrmm.math.uni-karlsruhe.de
Uni Karlsruhe, Institut f. wissenschaftliches Rechnen u. math. Modellbildung
Engesser Str. 6, 76128 Karlsruhe