8ms/99th write percentile latency - is it fast? (...)
summary
Claims taming self‑hosted Cassandra at 400k writes/sec and achieving ~8ms 99th‑percentile writes by applying SRE practices: Terraform for infra as code, Jenkins jobs to automate node replace/repair, a homegrown "spotty" spot‑instance deployment, and focused self‑healing automation wired into monitoring and runbooks. Recovery scenarios and chaos‑style workshops drive confidence; after 1.5 years the cluster stops paging and maintenance drops to 1 hour per two weeks, with 95th pct latencies around 2ms and operational runbooks proving effective.
AI-written summary · about
chapters (9)
- 0:00Introduction and reliability mindset
- 6:29Joining a no-ops company
- 13:58DevOps is not a role
- 20:03Reliability as a feature
- 27:36Understanding toil and automation
- 32:18Building reliability culture
- 36:11Working with business on SLOs
- 40:30Self-healing and Cassandra operations
- 43:27Achieving 8ms 99th percentile latency
related talks
48m8ms/99th write percentile latency - is it fast?
40mLatency, Throughput & Fault Tolerance: Building a Modern Database
39mMaking Cassandra Faster in Cloud-native architecture
35mReplacing Cassandra: A Digital Transformation for the World’s Largest Digital Identity Network
57mDSS Keynote — Back to the Future: Prepare Your Data Infrastructure for Anything
28m70 TB Servers – How and Why We Use Cassandra
27mReducing Client Latency and Timeouts When Running Cassandra on Public Cloud
33mCassandra Performance Tuning Like You've Been Doing It for Ten Years