Clustering
Now I have SMB humming on a Single machine my next step is to get the cluster going.
I found the doc http://docs.saffronmb.ws/How-to-Stand-up-an-SMB-Node doesn't go into the clusering config too much so was after a bit of advice/hand holding.
Can you give a rough guide on the steps required to get a cluster with SMB going?
Are there any specific setup settings that improve SMB's performance?
Do you use shared disk space just for ingesting data or does it need to be setup to do more than that?
Are there any gotcha's that I should be aware of?
Thanks in advance
Gareth
Comments are currently closed for this discussion. You can start a new one.
Support Staff 2 Posted by David E. Young on 31 Jul, 2009 01:09 PM
Hi Gareth. So by "clustering config" do you mean just telling SMB about the nodes in the cluster, or are you also asking about hardware and OS setup?
If it's just SMB configuration, I think the easiest approach, since you have SMB installed on a single node, is to follow these steps. I'm going to assume networking is set up properly on your cluster; that all nodes can talk with each other; that you can ssh into each node (as both saffron and root) without a password; etc.
We'll assume you've logged in as user saffron and are in the smb directory. Let's set up a two-node cluster for now.
I. Edit conf/SaffronCluster.xml. This file tells SMB what nodes are in the cluster and what their roles are. In your case you should have just one <node> entry (for the admin node). Replicate this line, adding it *after* the line for the admin node (it's a requirement that the first <node> is the admin node entry).
On the new line, give your new node an <id> different from the admin's <id>; the safest choice is to add 1000 to admin's <id> (i.e. if admin's <id> is 1000, use 2000 as the worker's <id>).
Next, change the <hostname> field to the name of your second node. The remaining fields should be set like this:
ClusterCoordinator="0"
Lookup="0"
Cache="0"
Later you can add lookup/cache services for improved performance.
Save your work.
II. Distribute SMB to your new node. Just run the rd command; this script will read your new SaffronCluster.xml and set up your cluster for a two-node SMB installation.
III. Now, hopefully you don't have any ingested data you want to keep. Assuming this answer is 'no', then run these next two commands:
creset -X (answer 'yes' to the prompt if you agree)
creset -f (answer 'yes' to the prompt if you agree)
IV. That should be it. You may now run 'cluster start' and your two-node cluster should come up.
Regarding your "shared disk space" question. When we ingest file-based data on our clusters here, we use the nfs setup as described in the SMB Node howto. If you're familiar with nfs and the automounter then you can alter that configuration to suit your needs. Otherwise, my recommendation for now would be to set up your nfs environment exactly as the howto describes, then place your data in the /opt/saffron/fbis directory on the head node. Your worker node will then have access.
Regarding performance tuning. I think either Yen-Min or Jim would be better qualified to give a detailed answer. However, the simplest change I tend to make on larger clusters is to add lookup services to worker nodes. You may do this in your SaffronCluster.xml; change the Lookup and Cache fields to "1" for the worker nodes you want, then do this:
rd
creset -f
again, if you have data you want to keep the process is much more involved. In that case we should talk by phone.
Hopefully this will help. Let us know if you need more assistance.
-- david
3 Posted by Gareth on 31 Jul, 2009 01:38 PM
Hi David,
Thanks for this. This is a great help, I'll keep you posted on how I get on.
Thanks,
Gareth
4 Posted by Gareth on 05 Aug, 2009 08:28 AM
Hi David,
The cluster is working fine now and the data was ingested successfully.
Thanks for your help
Gareth