[SGVLUG] Are there any oracle gurus on this list?

[Emerson, Tom]

> -----Original Message-----
> Behalf Of Max Clark
>
> ... There is not a _real_ database on the planet that you have to take 
> offline to backup.

heh heh heh -- this is one of the things I like about IMAGE on the HP -- there really isn't a concept of "taking the database offline" because there simply isn't any way to do it -- the "database" is nothing more than a set of system-protected files, and calls for accessing and manipulating the database are built in to the system, so if the SYSTEM is up, the database can be accessed.  (there's no "daemon" to start or stop)

OTOH, a process can access the database in an EXCLUSIVE mode, which is essentially what happens when the database is being stored, so in that way you could say the database is "offline", but that's only a convenient description.  In fact, you could write programs in such a way as to require "exclusive" access to the database and then not have to deal with "locking" of entries while you update them (but this isn't done all that often...)  While your process is running, the database would be "offline" to other processes, but "online" to the process that is updating it. (there is also an exclusive/shared-read mode where other process can read the database while a single process performs updates)

In the past, the process of storing a database would have it locked for the entire duration of the store process (which could and often would include other non-database files).  As the OS and utilities improved, a near-online and true-online backup process.  (though third-party providers are quick to point out that "true" only works if everyone plays nice -- a situation nearly impossible to guarantee)  The "near" online backup would essentially require about a 5-minute window either at the start or end of the store, to essentially create a synchronization point; the "true" online I think maintains a log of transactions that occured DURING the physical storage of the file(s) and includes that as part of the storeset.  [details here for the truly intrigued:
    http://docs.hp.com/en/B5151-90003/ch07s03.html 
note that they can apply this "shadow logging" to ALL files, regardless of whether or not they are database files]

I imagine these techniques are similar to what the "real_databases" employ [though having to start a process JUST to manage access?  come on, a "real" database doesn't need that, now does it? ;) ;) ;)]