Design a system,
Input: XML files containing transactions data
Output: JSON file, with each transaction object converted to json object.

Some stats:

1. File size - ~ 2GB
2. Files/second - around 100 files
3. The format of the xml will be simple. Transactions tag, containing all transcation objects.
4. Resulting json file can be output few minutes. No need of real-time oputput, or giving output json file in few seconds.

I was stumped on how to parallelise conversion of xml to json for a single file. Any ideas about that?
How would you go about clarifying this question further and how would you approach it.