Greetings,

I recently heard that Oracle 23 will support JSON! I am excited about this since I have been taken away from supporting Oracle for the past three years to work on Mongo.

As I have been working with Mongo, I’ve realized a few big weaknesses over RDBMS design. I wonder if Oracle's JSON implementation avoids these weaknesses:

1.) Storage overhead from repeating field names in each document.
a. Let's say I have a collection that will contain 1000 fields. In this collection we will have 1 billion documents. Each document will be required to add a value for each of the 1000 fields. I calculated the number of total characters used for the 1000 field names to be 10,000 characters. Does Oracle save the field names for the 1000 fields in each document. That will result in 10,000 characters times a billion documents = 10 Trillion characters stored! In relational database tables, only 10,000 characters would be stored for the table since they only store the field names in the table definition.
b. Do you know if Oracle’s JSON implementation won’t store the repeating field names in each document of the same collection? There are cases where we don’t need the “flexible schema” option. We want all documents to have the same fields and set data types.

2.) Mongo has a two replicaset requirement for each Mongo primary server instance. Is Oracle going to require the same, which is to have two standby database servers for each primary server?

3.) Mongo highly recommends that we have enough memory to store the whole index and collection working set in memory. The weakness here is that Mongo licensing is based on memory and Mongo's decision to store the field names per document significantly bloats memory with useless information. Will Oracle continue to license based off of CPU for JSON implementations?

Thanks,

John

create table t ( 
  jdata blob 
    check ( jdata is json format oson ) 
);