6

I'm having a small problem understanding Gluster Bricks and Nodes. In particular, on the 3.3.0 Administration Guide, eg, page 12 - 13, the discussion on "Striped Volumes" vs "Distributed Striped Volumes". See chapter 5, page 9, for the overview.

I'm looking to use it in a VMware ESXi 5.X environment for a HA NFS storage solution. As such the striped and distributed striped volumes are of interest as they relate to "high concurrency environments accessing very large files". This adequately describes Virtual Disks.

However, when reading the manual I have this:

  1. Striped volumes stripes data across bricks.
  2. Distributed Striped volumes stripe data across two or more nodes.

So I assumed a brick is (for example) a block device on a node (or a peer as they seem to refer to it sometimes). And a node is a distinct server (one of the peers).

However, the diagram on Striped Volumes shows server1 and server2 so that assumption seems wrong.

But then the distributed striped volume also shows the same thing, but differently. Now it shows server1 having two bricks (exp1 and exp2) which it calls 'striped volume 0' and then server2 also with two bricks (exp3 and exp4) called 'striped volume 1'.

The sample creation shows creating an 8 node 4 stripe volume which obviously would be different from the diagram having only 1 brick per node.

Also the example to create both types uses the same syntax.

Can somebody help with a clarification? What's a brick, and what's a node? Do I have it right and the diagram for 'striped' is in fact wrong, actually showing a distributed striped volume?

Patrick James McDougle
  • 207
  • 1
  • 9
Ian Macintosh
  • 945
  • 1
  • 6
  • 12
  • I think the diagrams are meant to be a configuration that is possible, but not the only possibility for that volume type. – Patrick James McDougle Mar 25 '13 at 16:31
  • For example, you could do the striped volume with 2 bricks on the 1 node. Or 2 nodes with 1 brick each. Or 4 nodes with 1 brick each. Or 1 node with 4 bricks. It just depends on how many places you want the glusterfsd running. – Patrick James McDougle Mar 25 '13 at 16:36

2 Answers2

11

A brick is any directory on an underlying disk filesystem. For example, you could create 4 "bricks" as /var/export/brick0 /var/export/brick1... etc., even though those directories were all on one disk partition. Not that we recommend this, of course, but it's possible. At its most basic, a brick is simply any filesystem you can export as a GlusterFS mount point. In that way, it's similar to an NFS directory export.

A node is a box in a trusted group of peers.

-John Mark Gluster Community Lead

JMW
  • 126
  • 2
  • You mention __recommendations__, what are those and where can we find those? Also, if you wouldn't mind, my post should probably be fact checked as I have only read, and never used GlusterFS. Let me know if any info in my post is inaccurate and I will edit it to fix. – Patrick James McDougle Mar 25 '13 at 21:07
2

The way I understand these things is as 3 separate ideas. You then combine them depending on your needs. Therefore you need to ask yourself 3 questions:

  1. If one of the bricks goes down (hardware failure) do I need the file to still be accessible? Highly available? If yes, then you need replication. NOTE: this is not a substitute for backups. If someone deletes a file, it is deleted everywhere the data is replicated. Retrieving this file would need to be from a backup.
  2. Will I need more storage space in the future? If yes, then you need distribution. This makes it so that as you need more space, everything gets spread out evenly over the number of bricks you have. Adding a new brick gives you more space, and re-distributes the existing files over all the bricks.
  3. Are the files really big, and a significant performance gain will be had by allowing files to have parallel access? If yes, you need striping. Striping spreads the file over different bricks for faster access.

You then combine your answers to these questions and decide on the volume that will work the best for you.

Patrick James McDougle
  • 207
  • 1
  • 9
  • Tks for comment Patrick. I already understand all those, but it's not the answer unfortunately. I'm looking for a glossary type clarification of 'brick' and 'node' as used in Gluster. – Ian Macintosh Mar 25 '13 at 15:15
  • I'd imagine that a node is a "server" or some entity running the glusterFS server daemon. And a brick is a volume on those nodes. A node can have multiple bricks, that is, multiple volumes on the same server. – Patrick James McDougle Mar 25 '13 at 16:24
  • Your comment above is correct Patrick given the clarification from John. It is what I originally assumed but got thrown by the diagram showing two nodes instead of two bricks on the same node. Unless my comment on John's answer is wrong :-) – Ian Macintosh Mar 26 '13 at 17:38