The Art of the Possible
summary
Demo shows Aerospike loading 10 million writes in ~20 seconds on a six‑node cluster (m6gd 2xlarge with 474 GB SSDs) and sustaining ~80,000 TPS with ~200 micros client roundtrip latency. Data placement uses a 496‑partition digest to decide a single network hop to the master node; servers keep a 64‑byte primary index in memory and flush small memory buffers to raw SSDs. Strong consistency is achieved with two copies via Aerospike’s roster protocol, surviving single‑node or rack failures.
AI-written summary · about
chapters (13)
- 0:00Aerospike introduction
- 1:24Why Aerospike is fast
- 3:09Aerospike architecture
- 6:23SSD leverage in Aerospike
- 7:25Efficient write handling
- 9:10Live demo setup
- 11:58Hardware and consistency
- 12:20Node failure simulation
- 14:04Recovery after failure
- 15:29Rack failure simulation
- 16:33Strong consistency with two copies
- 19:45Maintaining consistency
- 20:46Closing and resources
related talks
49mJourney with Aerospike: Architecture Best Practices
29mWhat breaks when Aerospike hits 6 million QPS
50mAerospike Technical Overview
24mCloud Workloads At Petabyte Scale
18mProduct Review and Roadmap
15mAWS On Air Voice of the Customer ft. Aerospike, Inc.
33mInnovation with Aerospike: How AppLovin Handles Billions of Requests per Day
32mFuture-Proofing your Architecture - AI at Any Scale with HPE & Aerospike