Cloud Workloads At Petabyte Scale
summary
Aerospike runs a hybrid memory architecture that stores indexes in DRAM or persistent memory and data on NVMe flash, compressing a 1 PB uncompressed dataset (500 billion 2 KB objects) to 250 TB per replica with a 4× compression factor and replicating across two AZs for 2 PB total. The benchmark places the cluster on twenty i3en instances (two clusters of ten) with c5n clients and reports 3–5M transactions/sec with sub-millisecond latency, demonstrating tight TCO on i3en.24xlarge storage-optimized nodes.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:45Trends
- 3:01Security
- 4:21Intel and AWS
- 5:46Intel Technology
- 7:43AWS
- 9:34Trade Desk
- 13:25Petabyte Database
- 14:26Instances
- 16:19Initial Database
- 17:21Summary
- 19:44Questions
related talks
15mAWS On Air Voice of the Customer ft. Aerospike, Inc.
45mAerospike + Persistent Memory = Voila!
22mThe Art of the Possible
25mCash the Cache - How BigBasket uses Aerospike
11mThe Power of Multi-Model: KV, Document, Graph
25mHybrid Memory Aerospike @ PayPal
28mAerospike's Journey Through Innovations, Sustainability, and Roadmap
49mJourney with Aerospike: Architecture Best Practices