Clustered NAS OS?

[dave_the_nerd]

Is there any linux distro geared towards making a clustered file server relatively easily? (like this or so). Some kind of uber-NAS4Free. Bundled packages, somewhat automated config process, etc.

Most of what I can find about doing it on Linux involved CentOS, GFS, and a LOT of manual fiddling, which I'm not really interested in since, well, on a schedule. (This is for work. I don't have to cluster it. Nobody's expecting that. I just think it's probably the Right Thing To Do so I'm trying to research my options.)

 

[VirtualLarry]

I think that such a thing would be pretty awesome. But wouldn't the client OS need to support a clustered FS?

 

[dave_the_nerd]

I think that such a thing would be pretty awesome. But wouldn't the client OS need to support a clustered FS?

No, just the NAS server nodes.

File Servers (NAS) serve files, using protocols like SMB or NFS. The underlying filesystem on the server's storage is transparent to the client OS.

 

[dave_the_nerd]

I wonder if I could just use DRDB and then "cheat" by having the backend SAN dedupe the LUNs...

hmmm...

 

[adamantine.me]

Isn't a NAS with a torrent service installed essentially just this? The data is provided from each node, it's just that the data is being transferred a different way compared to NFS/SMB/Rsync... Which reminds me, Rsync is also good at transferring only the missing data from a file.

Are you thinking something like a Beowulf cluster, except with a distributed network? There are some technologies like this being developed.

 

[XavierMace]

napp-it (Solaris based) does Async replication out of the box, HA Clustering with a 3rd party extension. Not sure what the pricing is on either though for business purposes.

 

[dave_the_nerd]

Isn't a NAS with a torrent service installed essentially just this? The data is provided from each node, it's just that the data is being transferred a different way compared to NFS/SMB/Rsync... Which reminds me, Rsync is also good at transferring only the missing data from a file.

Are you thinking something like a Beowulf cluster, except with a distributed network? There are some technologies like this obeing developed.

Not really. BitTorrent replicates files, but ideally you want something that allows multiple modes to access the same block devices without colliding. Like a Beowulf cluster.

The tech is in quite a few commercial has products, as well as available open source. I'm more hoping somebody had gone to the trouble of making setup a little idiot proof, but it doesn't look like it. At least not without a price tag.

 

[dave_the_nerd]

napp-it (Solaris based) does Async replication out of the box, HA Clustering with a 3rd party extension. Not sure what the pricing is on either though for business purposes.

Hmmm... That's worth looking at. Thanks.

 

[mv2devnull]

(I was about to mention cLVM, pNFS and iRODS, but they might not be so rosy or on the spot.)

How about Ceph?
http://ceph.com/resources/downloads/

 

[Fallen Kell]

You are better off with cLVM + GFS2 + cman + luci + ricci. All of these are available on RHEL, CentOS, and many other linux distros. Red Hat has released plenty of documentation on configuring a high available (HA) global-fs2 (GFS2) and steps you through the process.

Something like the following guide walks you through the basic steps:
https://bigthinkingapplied.com/creating-a-ha-cluster-with-red-hat-cluster-suite/

But read the Red Hat guide first:
https://access.redhat.com/documenta...e_Linux/6/html-single/Cluster_Administration/

 

[dave_the_nerd]

You are better off with cLVM + GFS2 + cman + luci + ricci. All of these are available on RHEL, CentOS, and many other linux distros. Red Hat has released plenty of documentation on configuring a high available (HA) global-fs2 (GFS2) and steps you through the process.

Is that basically the same as the Pacemaker docs? (It mentions cman, but not cLVM, luci, or ricci. In fact, I've not read anything about the last two there.) Is this one of those things where the same software is broken into different FOSS projects but is 99.7% the same?

Something like the following guide walks you through the basic steps:
https://bigthinkingapplied.com/creating-a-ha-cluster-with-red-hat-cluster-suite/

But read the Red Hat guide first:
https://access.redhat.com/documenta...e_Linux/6/html-single/Cluster_Administration/

Thanks - a little light weekend reading, lol.

Any opinion on using ZFS instead of GFS2?
https://github.com/ewwhite/zfs-ha/wiki

I'm tempted to stick with the known quantity, but a lot of the guys I work with are nosy BSD Programmers who will give me some shit for using anything other than BSD/ZFS, no matter what.

 

[Fallen Kell]

There are a lot of pieces across different FOSS projects for doing HA and clustered filesystems. Rici/Lucci are configuration systems which help you more easily build and manage the cluster.xml file used by cman. cLVM is not always specifically called that, but is simply the cluster aware version of LVM. Pacemaker seems to be just another package to do the same thing, creating the same xml file that Rici/Lucci do. The reason I like Rici/Lucci is that it is a web frontend/backend for the system, which has nicer pulldown lists and such for the various options/items when configuring them so that you don't have to constantly have the manual open.

Personally I love ZFS, but you really need the right hardware for it (specifically ECC RAM and a good SSD with lots of resilient/many-re-write capable memory for the ZIL device). But at the end of the day, ZFS is not clustered, which means that if the server goes down so does the access to the storage. You have to manually import the ZFS to another system or still use the cman+rici+lucci to create a HA service that moves the ZFS filesystems between two+ servers if/when one fails.

 

[dave_the_nerd]

Edit: Never mind. I suck at scripting.