In my previous team, we instead use the string "null" to represent NULL. Its pretty dumb but we haven't thought of any other way.
What I like about DynamoDB is that you can scale the table easily. You don't need a dedicated DBA to maintain your database.
For what I don't like
1. You pay for what IOPS you allocated
2. You can't share free IOPS to other tables
3. Getting a snapshot of the whole DB is impossible, DB backups are not transactional
4. Use EMR or DataPipeline for backups
5. If you reach your IOPS limit, you need to retry your writes/updates instead of delayed ack's. Others uses libraries that limits writes but its per server and doesn't account the free writes on other servers since the accounting is on the client level
What I like about DynamoDB is that you can scale the table easily. You don't need a dedicated DBA to maintain your database.
For what I don't like
1. You pay for what IOPS you allocated 2. You can't share free IOPS to other tables 3. Getting a snapshot of the whole DB is impossible, DB backups are not transactional 4. Use EMR or DataPipeline for backups 5. If you reach your IOPS limit, you need to retry your writes/updates instead of delayed ack's. Others uses libraries that limits writes but its per server and doesn't account the free writes on other servers since the accounting is on the client level