Cluster/SAN filesystems supported by SAP on Linux
nsoley at sympatico.ca
Fri Mar 18 20:04:03 MET 2005
At 11:39 AM 3/18/2005, Prieto, Jose Manuel wrote:
>I'm trying to figure out a way of sharing a central SAN-based SAP
>installation (DB+CI+AppServs) among a set of computers (nodes) in order to
>move SAP components around upon demand or when required (e.g. a node
>failure). Of course, there is a virtualization layer in between (logical
>IP address and so on).
>I knew of the existence of OCFS. But I can't use it because it only
>supports Oracle's files (datafiles, controlfiles, archives). The current
>stable version is not a general-purpose CFS. I also started to work with
>GFS but is seems that it's not supported by SAP yet (as you mentioned).
>Veritas CFS is not supported either.
I will second Marcus's urging to look into the SAP Adaptive Computing
Environment as this is pretty much exactly what it does,
no point in re-inventing the wheel.
There is lots of info at http://service.sap.com/adaptive
You do not need a CFS to accomplish this, and while it is one way to do it
I would tend to recommend against it for similar reasons to why I've
generally disliked raw files for databases.
What you do need is a system that makes each device or filesystem visible
to every possible server but concurrent filesystem access from multiple
nodes is not necessary, as long as you can unmount a filesystem from one
node and mount it on another that is sufficient for moving function between
nodes, any SAN storage solution can be configured to provide this, and it's
long been leveraged by failover solutions for high availability.
I know that SAP has been tested with IBM's SANFS filesystem on Linux with
DB2 and there are some other things that have been tested as part of the
ACC Compliance test process.
Now ACC will not provide automated very fast recovery from a failed node,
for that you still need some sort of failover software, IBM's Tivoli System
Automation for Linux or Steeleye Lifekeeper.
More information about the linux.general