This document discusses how using Heimdall Data's database proxy platform can reduce Amazon RDS costs by up to 50% without requiring any application changes. Key features of Heimdall Data include SQL traffic management, query caching, read/write splitting, and connection pooling. A case study is presented where a customer's RDS CPU usage was reduced from 100% to 3% through SQL offloading, resulting in 50% reduced RDS costs and months of avoided development work. Tables show examples of cost savings from 33-44% by migrating from multi-instance RDS deployments to Heimdall Data with Amazon ElastiCache. Heimdall Data is available on AWS Marketplace for a free
7. The Challenge:
• RDS CPU at 100%
Why we choose Heimdall Data:
• SQL offload via Caching and Read/Write split
• No application changes
Benefits:
• CPU reduced from 100% => 3%
• Cache hit rate: 90%
• Reduced RDS costs by 50%
• Saved months of development work
Customer Success Story
9. SQL Result caching and invalidation:
• No application changes
• For any Grid-cache (i.e. Elasticache, Redis, Hazelcast)
Uses real-time analysis and statistics on:
• Query frequency and variability
• Relative performance of Cache vs. Database
Provides:
• Auto-cache only if there is a performance benefit
• Cache recommendations and benefits
How Query Caching Works
10. Read/Write Splitting
Added value for AWS Customers:
1. Routes to Read/Write instances
2. Lag detection-aware: ensuring data retrieved is fresh
3. No application changes; configured by rules
Application
Server
Application
Heimdall
DB Proxy
Read 1
Read 2
Write
11. Putting it all together
Heimdall
DB ProxyApplication
SQL SQL
Amazon
ElastiCache
Heimdall
DB Proxy
Application
SQL SQL
Amazon RDS
Write
Amazon RDS
Read
13. Query Caching Use Case
55% Off Heimdall Data Promotion available until June 30th, 2021!!
Original Configuration vCores RAM Real Cost New Configuration vCoresRAM Annual Real Cost w/o Discount
Discoun
t
Real Cost w/
Discount
RDS SQL Server (r5.4xlarge)
MAZ 16 128 $ 120,140
RDS SQL Server (r5.2xlarge)
MAZ 8 64 $ 60,208 $ 60,208 0% $ 60,208
Total $ 120,140 Heimdall (r5.xlarge) x 2 4 32 $ 38,550 $ 38,550 55% $ 17,348
Elasticache (m5.xlarge) x 2 4 16 $ 1,741 $ 3,482 0% $ 3,482
Total $ 102,240 $ 81,038
Savings 15% 33%
14. Query Caching Use Case – 3 instances
Original Configuration vCores RAM Real Cost New Configuration vCores RAM Annual Real Cost w/o Discount Discount Real Cost w/ Discount
RDS SQL Server (r5.4xlarge) MAZ x
3 16 128 $ 360,420
RDS SQL Server (r5.2xlarge)
MAZ 8 64 $ 180,624 $ 180,624 0% $ 180,624
Total $ 360,420 Heimdall (r5.xlarge) x 2 4 32 $ 38,550 $ 38,550 55% $ 17,348
Elasticache (m5.xlarge) x 2 4 16 $ 1,741 $ 3,482 0% $ 3,482
Total $ 222,656 $ 201,454
Savings 38% 44%
55% Off Heimdall Data Promotion available until June 30th, 2021!!
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure
Each query is parsed to determine what tables are associated with it. The last write time to the table was tracked by Heimdall, and synchronized using Elasticache. This provided the first piece of information (i.e. when was the last time a particular table was written to). Next, Heimdall performed writes on a scheduled basis to the write node and read how long it took for the writes to appear on the read server. With this information, Heimdall intelligently decided on which node was “safe” to read from--either the read nodes, or only the write node.
Heimdall Data is:
The only solution to remove the app-database latency
Database vendor neutral, no vendor lock-in
Requires ZERO changes to your existing infrastructure