Persistency question...

[Cuong Nguyen]

Frank Wrote:
> Version 3.4 will be released until 31th March 2000. I have 
> already reworked
> the persistency concept. It's know significantly faster and 
> needs less disk
> space.
> A framework for a Load-On-Demand MibTable is rather 
> complicated, because
> it needs something like a database system below. Using the 
> file system for row
> 
> indexing seems a bit overkill for me.
> 
> As a consequence, I will implement an abstract 
> VirtualMibTable class, that
> implements all necessary interfaces needed to integrate a 
> real database into
> an AGENT++ table.
> 
> Any comments?

I think that makes sense.  If you would just provide the hooks we can
provide the integration into a database with out much problems.  I've
implemented an SNMP agent before where I wrap the SNMP stack around a set of
C++ classes which provided persistency via the GNU DB (gdbm).  The way I did
it was to "calculate" the key to the record using the OIDs of the rows.
This way searching for a particular OID was easy because GDBM provides the
necessary search for it.  The issue comes in when I have to implement
getnext.  For that I implement a search tree optimized for lexicalgraphical
walk and place the search tree into a seperate GDBM database.  I can now
index anywhere into the table and find the next logical OID.

I think this implementation could be general enough to make it work with
Agent++ and everything would still be open source (actually I got a bit
confused about the Agent++ licensing - I guess it is not GNU license so
maybe it doesn't matter if we try to stick to GNU license only software - in
that case there maybe other database options).

Cuong.