BerkeleyDB with SAN

Question

We have a use case where we are using BerekeleyDB as read only. As of now we are using local hard drive for accessing Berekely DB files.
Since the IOPS is very low with a local hard drive, we are having problems with scaling our appliaction. We have multiple nodes running our application.Since we need to repliacate the updated Berkeley DB files across nodes, it becomes a maintanence problem and have this overhead.
We are looking at a SAN solution to replace the local hardrive. which gives better IOPS and relieves us from overhead and maintaining files across multiple nodes.
Have few questions on this
1) I read in the FAQ secton that the behavious is undefined if SAN is used. That talks more abt when we have writes to the Berkeley DB. We are using Berkeley DB as a read only solution. Are they any issues we need to be aware of we we plan to use SAN?
2) It also talks about if we use SAN drives,we may not be able use file system cache? We are planning to use Berkeley DB cache and leverage Linux file system cache to reduce the IOPS. Is there an impact to that if we use SAN?
3) It talks about mutex locks while access Berekely DB. Can we avoid using locks, as we are using it as read only? Does the Berekely DB API provide any options to avoid locks on the data?
4) Is there anything else we need to be aware of while using SAN with Berkeley DB?
Appreaciate you could give some directions on the above questions.
Thanks

Answer 1

Hello,
It sounds like there is some part of the application which is not read only
since there is a need to replicate updated Berkeley DB files across nodes.
Even if the application is read only, if there are other concurrent updaters
accessing the database files, then all of them need to share the same environment.
If all instances of the application(s) are shut down whenever the user database files
are changed then the read-only nodes, each with their
own private environment and log files, could share access to the SAN database files.
Each node could use whatever BDB and Linux caching they like. Just be sure to do
something like mount the SAN database files read-only so there is no chance of
them being updated.
Mutexes are needed whenever a node could have more than one thread or process accessing
the environment.
You did not mention which product is being used. You may want to consider the
HA product or Concurrent Data Store product if transactions are not used.
http://download.oracle.com/docs/cd/E17076_02/html/programmer_reference/intro_products.html
Thank you,
Sandra