Building a Telecom-Scale OLAP Platform with Apache Doris
Автор: VeloDB
Загружено: 2025-11-14
Просмотров: 104
As part of this talk, we plan to walk the community through how we have designed and scaled Apache Doris as the central OLAP backbone for Onextel, currently handling 350–400 million SMS events per day, and our roadmap to scale further. The session will cover the following key aspects:
Business Context
1. The need for a unified, high-performance analytics platform capable of ingesting and querying hundreds of millions of daily events.
2. Why we selected Doris over other OLAP systems for this mission-critical workload.
Data Modeling at Scale
1. Implementing a fact-dimension model tailored for multi-tenant, high-cardinality data.
2. Using Duplicate Key tables for Kafka metadata (late-arrival handling and replay management).
3. Using Unique Key tables for fact data (ensuring correctness via upserts and de-duplication at scale).
4. Using Aggregate Key tables for dimension/summary rollups to power fast analytics.
Performance Engineering
1. Tuning Doris using:
a. Tailored compaction strategies to minimize write amplification
b. Data-statistics-driven indexes (inverted, bloom filters) to accelerate query performance
c. Partitioning and concurrency optimization to support high-volume adhoc queries with low latency.
Reliability & Operations
1. Prometheus + Grafana dashboards built on Doris-exposed metrics to monitor:
a. Routine-load health, ingestion lags, and throughput
b. Query concurrency, BE node CPU/memory utilization
c. Incident handling for ingestion stalls, node failures, and compaction issues.
Telecom-Grade Reporting Layer
1. Building a custom reporting API layer on Doris that powers:
a. Real-time delivery analytics
b. Tenant-wise billing dashboards
c. Long-tail ad-hoc querying for operational teams
AI Use Cases
1. Leveraging Doris as the data foundation for MCP-driven AI workflows, including:
a. LLM-powered operational insights on messaging data.
Future Roadmap
1. Expanding AI use cases on Doris datasets
2. Data cataloging and governance via Apache Iceberg + Unity Catalog
3. Transitioning to a Lakehouse-style unified architecture with Doris as the core serving layer
4. Evolving the platform to seamlessly handle 1 billion+ daily events while maintaining low-latency analytics.
#apachedoris #telecom #onextel
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: