[OpenAFS] AFS design question: implementing AFS over a highly-distributed, low-bandwidth network

Chaz Chandler clc31@inbox.com
Mon, 19 Jan 2009 10:33:08 -0800


Thanks for everyone's replies=21

So far, in summary:
1) There should be a separate cell at each site.  This avoids quorum and =
and local client server selection difficulties that arise in a VPN =
environment.
2) There is no good AFS-based solution for group shares in this scenario.
3) All volumes that need to be read from or written to at LAN speeds at =
multiple sites must be replicated via custom scripts.
4) Disconnected operations would be nice, but are currently not =
implemented at a level of completeness suitable for a production =
environment.  Traveling users will simply have to access AFS through the =
VPN and deal with the slow speed=21

Further questions:
a) What is the best way to replicate a volume across cells?
b) How would the presence of multiple cells effect the krb5 infrastructure =
(currently: one realm, one cell, cell name =3D realm name =3D internal LAN =
domain name)?
c) Are any of the Morgan Stanley volume management system utilities =
available publicly, or are their methods sufficiently documented publicly? =
 All of what I've read about them are from previous afsbpw's. (ie, =
http://workshop.openafs.org/afsbpw08/talks/wed_1/OpenAFS_and_the_Dawn_of_a_=
New_Era.pdf)

-Chaz

> -----Original Message-----=20
>>>> Something that might work out a little better is the work being done
>>>> on disconnected operation.  That might suffice for some of your use
>>>> cases (assuming the timing of that finishing and the features it
>>>> will
>>>> offer is suitable for your needs).
>>>>=20
>>>=20
>>> Is disconnected afs functional?  I saw it as a GSoC '08 project but
>>> the afsbpw08 presentation said it's only R/O.
>=20
> The BPW presentation predates the work that Dragos did as part of
> Summer of Code. His project was very successful, and delivered a
> working disconnected client implementation which is now in the 1.5.x
> tree for testing.
>=20
>> I would not inflict it upon any of my power users or regular users.
>> R/W
>> does work, but rebooting will lose data and it doesn't work on
>> windows,
>> just Unix. Simon can tell you more.
>=20
> There's a couple of bits of major functionality missing, and some bugs
> which still need to be ironed out. In the functionality stakes, we
> currently don't preserve some important information when the machine
> is rebooted, so you lose any changes made whilst disconnected if you
> restart the machine without reconnecting. Less seriously, but still an
> issue, is that there's no mechanism for telling the cache manager
> which files you'd like to have available whilst disconnected. At
> present, you need to explicitly access every file and directory you
> require before disconnecting.
>=20
> We're also still in bug hunting mode. There are some bugs with the
> disconnected implementation which are being found and fixed, and there
> are some race conditions in the cache manager which are being exposed
> by the differences in timing and serialisation when you are
> disconnected.
>=20
> Cheers,
>=20
> Simon.

____________________________________________________________
Receive Notifications of Incoming Messages
Easily monitor multiple email accounts & access them with a click.
Visit http://www.inbox.com/notifier and check it out=21