AI 용어집
인공지능 완전 사전
Retention Policy
Rule defining the automatic retention period of time-series data based on their age, allowing for the deletion or gradual archiving of old data to manage storage space.
Data Rollup
Process of automatically aggregating raw time-series data into precomputed metrics (averages, sums, minimums) over defined time intervals to accelerate analytical queries.
InfluxDB Line Protocol
Text format optimized for high-performance ingestion of time-series data, structuring measurements with tags, fields, and timestamps in a single text line.
Chunking Strategy
Method of dividing time-series data into segments (chunks) of defined size or duration to optimize read/write operations and compression of time series.
Tag-Value Optimization
Technique involving the use of indexed tags for dimensional metadata rather than fields, enabling much faster filtering queries on large volumes of data.
WAL (Write-Ahead Log)
Sequential transaction log where time-series writes are recorded before being applied to the main database, ensuring data durability in case of a crash.
Series Encoding
Time-series specific compression method exploiting temporal patterns (delta encoding, gorilla compression) to drastically reduce required storage space.
Query PromQL
Functional query language specifically designed for time-series data, enabling complex selections, filtering, and aggregations on multi-dimensional metrics.
Prometheus Remote Write
Standardized protocol for sending time-series metrics from Prometheus agents to centralized storage systems, supporting high-frequency streaming writes.
TSMetric Ingestion
Process of high-speed time-series metric ingestion, involving parsing, validation, indexing, and storage optimized to handle millions of points per second.
Temporal Sharding
Strategy for partitioning time-series data based on time, distributed across different nodes or servers to balance load and horizontally scale performance.
Backfilling Strategy
Approach for retroactively populating a time-series database with historical data, requiring specific optimizations to avoid impacting real-time performance.
Time-Series Indexing
Hybrid indexing structure combining tag indexes and temporal indexes to enable ultra-fast searches across trillions of time-series data points.
Gorilla Compression
Compression algorithm optimized for time-series data using delta-of-delta encoding for timestamps and XOR for values, achieving 10:1 ratios while preserving precision.
High-Frequency Sampling
Collection of time-series data at very short intervals (sub-second), requiring specific optimizations to handle massive volumes while maintaining low latency.