I would look at what Oracle is doing with 11i, and try to replicate whatever configuration they have done most testing with, whatever they recommend as a best practice.
IMHO automating patch installation, and fast replication of installations between systems is an easier class of problem to solve than making a shared filesystem dependency reliable. Have you looked at ZFS snapshot replication? With compression and decent network bandwidth it should be possible to replicate the common parts of the Oracle installation directory rapidly, and after a patch it only has to replicate the changes from the previous snapshot. For cost savings, put the Oracle installation on mirrored local disks, its faster, cheaper, more reliable (especially with ZFS) and its essentially read-only, so no need to have it backed up or loaded on the EMC. Adrian On 11/14/07, Hameed, Amir <[EMAIL PROTECTED]> wrote: > Adrian, > Thanks for the reply. Currently we are looking at the shared versus > non-shared options and once we weigh the pros and cons then we will > decide which way to go. With the Shared Application Filesystem, we will > achieve the following benefits: > - Any Oracle application patch will need to be applied only once and > from any tier. This is one of the main issues that this architecture > will address. In a multi-tiered, multi-node (Oracle Apps) architecture, > if we use non-shared storage then we will need to patch each an every > node. Just to give you an idea that currently we have an existing > mission critical system that is comprised of a backend database server > and two middle-tier web servers using non-shared storage. During large > releases where it is not feasible to patch both the middle-tiers due to > the time constraints, we patch the backend and one middle-tier server > and then replicate the second middle-tier server from the first and it > takes over five hours to do that. This way we will eliminate that. > - Storage savings: We use EMC standard and BCV devices. This will cut > down the storage cost. We are not talking just one environment here; > this will be implemented in the entire environment stream. > - Inability to use some of Oracle's own technologies, specific to the > 11i apps, where the shared architecture is required. > > So, there benefits are there. It is the technology that we are trying to > figure out to use to achieve this. > > Thahks > Amir > -----Original Message----- > From: adrian cockcroft [mailto:[EMAIL PROTECTED] > Sent: Wednesday, November 14, 2007 11:37 AM > To: Mike Gerdts > Cc: Hameed, Amir; perf-discuss@opensolaris.org > Subject: Re: [perf-discuss] NFS configuration over private network > > I think you need to stand back from the problem and figure out what you > are trying to achieve by sharing the Oracle installation, and whether > that benefit outweighs the complexity of the solution needed to get > there. > > For mission critical installations, minimizing the number of single > points of failure is most important. The extra work to install Oracle on > each node is offset by less exposure to problems, less chance that you > lose the entire RAC rather than one node, a more standard installation, > and much less work to clean up the mess when something does go wrong. > > My 2c... > Adrian > > On 11/14/07, Mike Gerdts <[EMAIL PROTECTED]> wrote: > > On Nov 14, 2007 8:42 AM, Hameed, Amir <[EMAIL PROTECTED]> wrote: > > > Mike, > > > Thanks for the valuable feedback. To provide high availability, we > > > will RAC this system. I would like to ask some follow up questions > > > and because I am not a system/Unix administrator (I am an Oracle > > > Applications and DB Engineer) so please bear with me if I ask a > > > question that does not make much sense. > > > - Because NAS also uses NFS, therefore, NAS is as secure/unsecured > > > as the OS-based NFS is? > > > > Exactly. > > > > > - In your opinion, the NFS-based architecture that I am > > > contemplating where > > > a) the database server will act as the NFS server > > > b) filesystems will be shared over a gigabit network > > > c)IPMP will be used to facilitate NIC failover is as robust > > > > and reliable as what any NAS appliance would offer. > > > > Assuming that you use VCS or Solaris Cluster to provide HA-NFS > > service, I would say that they are comparable. > > > > When considering performance, it is important to consider > > optimizations that exist in NAS appliances - it is common for them to > > have NVRAM to journal writes that allows them to offer better NFS > > performance. On the other hand, I have generally found it harder to > > debug any problems that come up with a NAS appliance is in the mix. > > Hopefully the OpenSolaris emphasis on becoming a storage platform will > > > change the industry for the better in this area. > > > > -- > > Mike Gerdts > > http://mgerdts.blogspot.com/ > > _______________________________________________ > > perf-discuss mailing list > > perf-discuss@opensolaris.org > > > _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
