On Wed, 2008-08-20 at 16:16 -0400, Daniel Davis wrote:
> I am concerned about using NFS for Fedora's mount point. I need to
> check with Chris Wilper because I seem to recall some problems when
> the network goes flaky (not to mention that NFS is a security mess
> unless it is behind a well controlled firewall). Plus NFS is not
> really fast. NFS is slowing dying off in most data centers except for
> backup or very fault tolerant cheap NAS storage. My tests were with
> high performance connections such as dedicated direct connection
> storage networks, iSCSI, and Fibrechannel. Unless you use a
> connection that runs close to direct connection speeds you will really
> slow down disseminations (and ingest).
I was thinking more along the lines of iSCSI or fiber, I just said NFS
for dev work, but those are good points, NFS is pretty much unneeded esp
with storage sizes today.
> The new Akubra project will likely take up how to store over less
> reliable networks.
> ALL the bitstreams and all the FOXML are stored as files. This lets
> you rebuild if the DBMS or Triplestore fails. The DBMS (and the
> Triplestore if you use one) is there to speed up operations or enable
> queries. Only a very limited amount of data is stored in the DBMS
> (metadata kind of stuff extracted from the FOXML).
This is what I thought, but didn't completely understand, thanks for the
> It is perfectly OK to have the DBMS and FC run on the same server in
> all but the very largest installations. Most installations run in
> your proposed configuration. In a very large repository you may
> consider running the DBMS on its own server which is tuned just for
> that purpose.
> While you can store just the bitstreams on your SAN/Fabric, I would
> keep the objects there too. If you have a well performing SAN/Fabric
> connection you can store everything on it except I would keep the
> operating system on a locally attached disk for bootstrap convenience
> and the swap partition.
This is the arch I'm going to setup then, this is great, I appreciate
it! This setup won't be huge (70k digital objects) and will focus on
the back end for archival and a base for harvesters to hit.
> -- Dan
> Phil Cryer wrote:
> > On Wed, 2008-08-20 at 14:57 -0400, Daniel Davis wrote:
> > > It should work fine. I have set the mount point within a Brocade Fabric
> > > partition and it worked perfectly. The one thing to remember is that
> > > only one Fedora instance should control the files. It is OK if the
> > > mount point is shared (though you may lose performance) but the specific
> > > storage directories must not be shared between two Fedora instances. I
> > > have had no problems using a well configured SAN/Fabric for DBMS or
> > > Triplestore persistence. It is generally better to dedicate the
> > > partition to a specific server and not share it when using it for
> > > Fedora, DBMS or Triplestore persistence.
> > >
> > Thanks for the reply - so would I be able to run the mount point over a
> > NFS connection to the SAN, or would it have to be hard connected? I was
> > thinking of having the DBMS on the same server as FC, and then utilize
> > the NFS/SAN mount as the storage for the digital objects in place of the
> > way I currently do in FEDORA_HOME/data - is this right? Are the objects
> > (jpgs, sid images in this case) stored within the RDMS or are they just
> > on the filesystem in the data/objects directory, or am I missing the
> > basics on how this functions?
> > Thanks
> > P
> > > I you dedicate the partition you still have a great failover mechanism.
> > > Should the server die you can fail over to a second server and let that
> > > server grab the SAN/Fabric partition. You can run Fedora without a
> > > rebuild if the failure of the first server was clean with respect to the
> > > file store, database, and triplestore. Otherwise you can run the
> > > rebuilder but still gain an advantage in that the content bitstreams and
> > > foxml files need not be moved.
> > >
> > > Finally, you can cluster the DBMS if you are using one with that
> > > capability and/or use Fedora Journaling if you want a more loosely
> > > coupled installation.
> > >
> > > -- Dan Davis
> > >
> > > Phil Cryer wrote:
> > >
> > > > Is there any documentation about having Fedora-commons store objects to
> > > > a distributed filesystem - HDFS or the like, or just a local SAN? I
> > > > spoke with Sandy last week at RIRI about an S3 module, but I'm wondering
> > > > if there's something to map to a local DFS/SAN for now. Would it be a
> > > > case of just remapping FEDORA_HOME/data to mount the remote storage?
> > > > Does FC even have to know about it, or is it system level at that point?
> > > > Does it have something to do with:
> > > > <param name="file_system"
> > > > value="fedora.server.storage.lowlevel.GenericFileSystem">
> > > >
> > > > Thanks
> > > >
> > > > P
> > > >
> > > >
> Daniel W. Davis
> Chief Software Architect, Fedora Commons
> Researcher, Cornell Information Science
> (607) 255-6090 (Office)
Phil Cryer | Open Source Development | Missouri Botanical Garden
www: http://mobot.org | latitude, longitude: 38.613877, -90.257943
email: phildotcryeratmobotdotorg | im googletalk/skype: phil.cryer