Redhat 6.2 Without Cluster - Shared LUNS - READ,WRITE

[Akif_1]

Hi All Experts!

#  I have two redhat nodes

# A shared LUN presented on both the nodes and mounted

# How can i make this two nodes read,write without cluster active,active.

?? Need expert advice to solve this issue.

Thanks & Regards,

[Matti_Kurkela]

Sorry, but I don't think this is possible in RHEL 6.2 without setting up a cluster.

Or, technically, you can just start using it... but you'll find that a regular non-cluster filesystem will get corrupted as soon as you perform a few write operations from both hosts. The non-cluster filesystems are designed to assume that only one host at a time will have access to the filesystem. This allows the filesystem driver to cache things efficiently: the filesystem driver can read something and trust that the data on the disk won't change on its own unless the driver actually receives an write operation to change it.

When you have two hosts mounting the same LUN without using a cluster filesystem, this design assumption will not be true. As soon as the filesystem notices that some essential filesystem metadata has changed on the disk (because the other host has changed it), the driver will output serious errors and normally will switch the filesystem to read-only mode. After that, a filesystem check is needed to evaluate the actual state of the filesystem and fix the inconsistencies in the filesystem metadata.

There are special cluster filesystems like GFS and OCFS2 that are designed to work correctly when accessed by multiple hosts simultaneously. But they have some requirements that non-cluster filesystems don't have: mainly, they need a DLM (Distributed Lock Manager) system running... and that requires having a cluster configured, because if one of your member hosts crashes or becomes unreachable while it's holding a DLM lock for a critical part of the filesystem, you will need some way to kick that host out so that the other host may safely either complete or rollback the operation the unreachable host was doing and then continue using the filesystem.

[Akif_1]

Hi Mati,

Iam really impressed with your honourable contribution in my real knowledge.

# If i planned to cluster this ( Hope license issue will not be a barrier)

# Plan is to use f5 load balancer and point the users on both the shared LUN /rersource FS as FILO/FIFO method.

# Guide me the best solution with docs to READ,WRITE shared lun on both the nodes and work as a failover.

Thanks a lot.

Regards,

[Akif_1]

Matti_Kurkela :Hope everything going fi9, I appreciate your valuable contribution and awaiting for my last reply.

Thanks & Regards,

[Matti_Kurkela]

I'm afraid this thread does not have enough information for finding the "best" solution.

You mention load balancing - is this about HTTP/HTTPS traffic, or something else?

If the traffic is HTTP or HTTPS, what kind of stuff is to be stored on the shared LUN? Are the clients just going to download information from it, or are they submitting some information that needs to be stored? How much? Would the clients be submitting small or large pieces of data? (i.e. individual records or entire files?)

What would the usage pattern of the shared LUN be like: would it be mostly reading, about equal reads/writes, or mostly writing?

Do you have any NAS available, either as a function of your current storage system or as a separate storage system?

If you have a NAS that is highly-available (fault-tolerant) enough for your purposes, the simplest solution might be to create a NFS or CIFS share on the NAS and mount it on both nodes.

If you must have a shared SAN LUN, and must have it active on both nodes, you might have to pay for the appropriate license for a cluster filesystem. But you should understand the strengths and weaknesses of your cluster filesystem before buying the license: for example, if your access pattern involves frequently creating and deleting small files and accessing them near-simultaneously from both nodes, that was about the worst possible use case for the original GFS. I hear that the current GFS2 is supposed to be better, but it is still something that you should understand before deciding on your system architecture. OCFS2, another cluster filesystem that you can license from Oracle as an add-on to RHEL 6.x, has different strong and weak points.

If your clients are frequently producing small pieces of data that must be stored and be accessible by both nodes, then I believe it might be beneficial to add another tier to your architecture: the server nodes would just have program binaries and possibly static content, and the client-produced and other dynamic data should be stored in a database backend. The database can then be made Highly Available using database-specific methods (e.g. MySQL cluster, Oracle RAC, or whatever) if so desired. Of course, this would be a major change to your software architecture...