Improve SQL scale and reduce RDS costs up to 50%: No code changes.

Amazon RDS video: Read/Write split, Query caching, & Connection Pooling


RDS Video Topic Video Start Time (sec)
Query Caching 9:20
Read /Write Split 13:25 
Connection Pooling 18.00 and 25:22
Q&A 18:25


Read/Write Split with Strong Consistency

The Heimdall proxy intelligently routes queries to the appropriate database instances without code changes. We load balance across replicas for better utilization, avoiding DNS issues.  Users receive the most up-to-date  data with our replication lag detection. Check out our AWS blog.

Heimdall for ElastiCache

Automated Query Caching: Removes latency to the database. Result sets are cached to the storage of your choice: 1) Local memory, 2) Amazon ElastiCache, or 3) Both. Automated invalidation is supported. Deployment requires zero code changes.

Connection Pooling

The Heimdall proxy supports true multi-user pooling. Operators can safely scale without overwhelming database resources. Techniques that reduce connection overhead include:

  • Connection Pooling: Multiple client connections are associated to a database connection. Users can limit
  • connections to the backend in 2-tiers: per user and per database. 
  • Connection Multiplexing: An extension of pooling, the proxy dispatches individual queries or transactions from the connection pool. As client connections are often idle, multiplexing creates a backend connection for “active” queries. The net result is lower total memory, and reduced database costs.

Connection Pooling and LDAP/Active Directory Video


Transparent Batching of INSERT’s

Heimdall Data improves performance over 1000x! We batch singleton INSERT’s under a single transaction.

  • Ideal use case:  INSERT a large amount of data at once on a thread. Heimdall processes at once much faster than if individual queries outside of a transaction were completed. See below demo video.


  • Not so ideal use case: Concurrent writes and reads against the same table, on the same thread will not be beneficial with this solution, as everything will just block until the DML operation is completed anyway.

Watch how batching INSERTs improved performance 1000x!


Redshift Video Topic Video Start Time (sec)
Batch Processing 0:00
Query Caching 7:30
Connection Pooling 12:25
Q&A 14:57


Automated Persistent Connection Failover

Our proxy detects database health and performs a failover to the standby. But doesn’t Amazon RDS already support this? How are we different?

The Heimdall proxy failover takes an application-centric approach: Upon a failure, we queue up the connection and transparently creates a new connection at the backend to the standby instance. This greatly reduces and/or eliminates application errors and exceptions. Hence, failover is transparent to the application and user; not so, with a DNS or IP based failover solution.