Hello,

I would like to know if GFS can be used without clustering.
That is, I would like to use GFS only as a file system to have multiple hosts access some shared file system. GFS probably has a lock manager that guarantees mutual exclusion between readers and writers. However I don't want to use the clustering capabilities of Linux (Red Hat Cluster Suite).

Also I would like to know if GFS allows multiple readers / single writer mode of operation. This mode of operation is much less expensive than multiple readers / multiple writers mode of operation.

Thank you in advance,
Guy

GFS can't be used without the setup of a cluster (the configuration basically tells GFS what to do). However, and that is what you meant in the first place, you don't need to configure Resource Group Manager (i.e. Cluster Suite, rgmanager).

You may mount the GFS filesystem with ro on the Reader Machines and you even may disable locking on the writer machine. Successfully done for example at a major notebook manufacturer who quickly wants to push the operating system (that resides on GFS and is writeable from a single server) to lots of laptops at the same time.

Thanks Brian,

I looked at Red Hat documentation on Cluster Suite and GFS. Indeed GFS requires Cluster Suite to be configured. This is how I guess the distributed lock manager (DLM) can be assigned to an active node in the cluster. But I only want the cluster software to manage the DLM used by GFS, nothing else. I would not like to see the cluster software to control where user traffic is processed. We use a load sharing solution where traffic is shared between two active nodes. If a node fails, the load balancer will take care of rerouting all traffic to the other node. I don't want the cluster software to do this. Maybe this could be done by having some dummy fencing configuration? Does this make sense?

Thank you again,
Guy

As GFS is used so that multiple nodes may access a shared storage device simultanously with concurrent read/write access, there must be a cluster wide lock manager who tells all nodes which files are currently locked by certain cluster members. This is done nowadays with the distributed lock manager (DLM).

If you use a loadbalancer to direct requests to specific nodes, the node gets to process the traffic, however for each request it still has to ask the lock manager whether it is safe to read/write the file at the moment and it has to wait until it gets the go.

In some cases you are able to disable locking at all but you need to be very careful as this may corrupt your data very easily ("-o lockproto=none" mount option).

I'm a bit surprised that you want to use GFS without fencing devices. Fencing devices are used for cluster safety to immediately block nodes from the cluster if there is an error.

Usually you would use more than two fencing devices besides the fence_manual option. They are mandatory for productive environments.