C++: Managing in-memory objects in relational database

Question

I'm currently facing a design problem. I'm writing an application that manages "documents" in a graph.. In a graph, each node has documents in its queue.

I finaly chose MySQL to manage the database and to persist the documents. The schema is mainly composed of 2 important tables: document and queueelement. queueelement has a reference to a document. A node's queue is a list of queueelement.

Basically, the classes look like the followings:

class Document
{
    int id;
    string name;
    //... 
    // other attributes here
    //...

    std::map<string, string> attributes;

    void foo();
    void what();


};

class QueueElement 
{
    int id;
    int docid;
    int priority;
    string nodename;
};

class Node 
{
    std::list<QueueElementPtr> elements;
};

The database schema is quiet simple and I don't need something more complicated.

A server app is managing the nodes. Basically, when a node is awake, it pops the first document in the queue, processes it and pushes it in a child node's queue.

I chose MySQL because I don't need a bigger database like Oracle and I won't need complicated sql queries. I also chose MySQL because I can't afford an OODB like Versant or Objectivity. Indeed, object-mapping is the key. I used Versant DB in trial mode, great product...but really too damn expensive.

I am thinking about how to load efficiently my graph in memory. In one hand, I am afraid to run into an Out-Of-Memory situation. On the other hand, I wish to avoid too many "select" operations.

I'm obliged to have the queues loaded correctly in memory. However I think that I'm not obliged to keep the "Document" object in memory, maybe loading it with a "select" when the node will process it would be enough.

Actually I am not a DB expert. I know SQL and how to create tables with respective indexes. But I must admit I ignore how the engine optimizes the I/O and the memory side. In my application, performace is a priority.

This application should run on a dedicated server and the MySQL database is located on the same server and will not host any other schema. I'm using MySQL c++ connector API, and each query (insert, delete, update) is predefined with a PreparedStatement object when the engine is initialized.

How would you manage a custom in-memory database using products like MySQL ? Do you think I'm taking the wrong direction ?

Thanks.

Z.