[DRBD-user] Peer FS? Anyone consider or try this?

Paul G paul at rusko.us
Tue Nov 23 21:56:51 CET 2004

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


----- Original Message ----- 
From: "David" <davidu at gmail.com>
To: <drbd-user at linbit.com>; "Paul G" <paul at rusko.us>
Sent: Tuesday, November 23, 2004 3:36 PM
Subject: Re: [DRBD-user] Peer FS? Anyone consider or try this?

> On Tue, 23 Nov 2004 15:23:53 -0500, Paul G <paul at rusko.us> wrote:
> > david,
> >
> > it's much more like afs or intermezzo than drbd. in fact, there's a good
> > chance it's based on them.
> >
> are you sure?  It seems like it works in a disconnected mode...and it
> doesn't grow your filesystem size by adding more nodes...it just
> replicates.
> Maybe I am reading wrong....

i didn't say it was identical, but rather *similar*. drbd, as you know,
works at the block device level. the guarantees that allows drbd to make are
important for some applications (ie databases) and not so important for
others. the peerfs guys wrote a filter fs. their "technical brief" is, to
put it politely, "interesting" reading as they can't seem to decide whether
they are consistent (coherency with 0 time bound) or "near-real time"
(coherency). the fact that they don't block until the rpc succeeds speaks
for itself (it's basically equivalent to drbd protocol A). needless to say,
this does *not* allow them to claim consistency and makes their coherency
claim a bit shaky as well (it is not impossible that an fs op would succeed
locally but not remotely).

by all means play with it and report on the results, it would be interesting
to see if nothing else. if i were testing it, i'd try to break metadata
stuff first.


More information about the drbd-user mailing list