HADOOP IN THE BIG DATA ECOSYSTEM

HADOOP IN THE BIG DATA ECOSYSTEM

1. HADOOP CORE ARCHITECTURE

   • Hadoop Distributed File System
     • This is Hadoop's storage layer, responsible for saving massive files across multiple machines.
       • Blocks
         • Files are divided into fixed-size blocks (usually 128 MB).
         • Block Size Tuning
           • Larger block sizes improve sequential processing performance.
         • Block Independence
           • Each block can be processed separately on different nodes.
       • Replication Factor
         • Each block is copied to several machines (default = 3) to avoid data loss.
           • Replication Policy
             • Data is replicated across nodes and racks.
           • Trade-offs
             • Higher replication = more reliability but more storage usage.
       • NameNode
         • Controls the file system metadata:
           • Which file is in which block
           • Where each block is stored
         • File System Journal
           • Logs all operations on metadata.
         • Vulnerability
           • If the NameNode fails, the system can become inaccessible.
       • DataNode
         • Stores the actual blocks and sends heartbeat messages to the NameNode.
           • Block Reports
             • Periodically sends reports with all blocks stored locally.
           • Local Storage
             • Uses the node’s local disk (ext4/XFS).
           • Failure Detection
             • If a DataNode fails, NameNode automatically re-replicates missing blocks.

2. HADOOP ECOSYSTEM

   • Apache Hive
     • Metastore
       • Stores table metadata (schemas, partitions, locations).
       • Maintains schema information.
       • Tracks table locations and partitions.
     • Execution Engine
       • Runs queries using MapReduce, Tez, or Spark.
     • HiveQL
       • SQL-like query language (Schema-on-read).
       • Supports joins, filters, aggregations.
       • Uses schema-on-read.
     • Hive
       • Data warehouse system on Hadoop.
   • Apache Pig
     • Pig Latin
       • High-level scripting language for data flows.
       • Easy-to-read transformation steps.
     • Parser
       • Validates and analyzes the script.
       • Builds the logical plan.
       • Detects semantic errors.
     • Optimizer
       • Improves the logical execution plan.
       • Improves execution efficiency.
     • Execution Engine
       • Executes dataflows (MapReduce, Tez, or Spark).
       • Converts logical plan into physical operations.
       • Manages parallel tasks.
   • Kafka
     • Distributed streaming platform.
       • Topics
         • Logical categories for messages.
         • Allow scalable data streaming.
         • Organize events by type.
       • Partitions
         • Enable horizontal scaling.
         • Support parallel consumers.
         • Maintain ordered logs per partition.
       • Producers & Consumers
         • Producers write data streams.
         • Consumers read messages.
         • Support consumer groups for load sharing.
   • Flume
     • Distributed data ingestion framework.
       • Sources
         • Support Syslog, HTTP, Netcat.
         • Capture high-throughput streams.
         • Receive incoming event data.
       • Channels
         • Temporary buffer storage.
         • Memory or file-based.
         • Guarantee reliable delivery.
       • Sinks
         • Send data to HDFS, Kafka, ElasticSearch.
         • Support batch delivery.
         • Handle failover paths.

3. APACHE SPARK INTEGRATION

   • Spark Core
     • Main engine for distributed computing.
       • RDDs (Resilient Distributed Datasets)
         • Immutable distributed data collections.
         • Support fault tolerance.
         • Enable parallel operations.
       • Transformations
         • Create new RDDs from existing ones.
         • Lazy evaluation for efficiency.
         • Examples: map, filter, flatMap.
       • Actions
         • Trigger execution of transformations.
         • Return final results.
         • Examples: collect, count, reduce.
   • Spark Modules
     • Specialized libraries built on Spark Core.
       • Spark SQL
         • Structured data processing.
         • Uses Catalyst optimizer.
         • Supports DataFrames and SQL queries.
       • MLlib
         • Machine learning library.
         • Supports classification and clustering.
         • Includes model evaluation tools.
       • GraphX
         • Graph processing library.
         • Supports Pregel API.
         • Enables graph computation and analysis.
       • Spark Streaming
         • Real-time data processing.
         • Uses micro-batching.
         • Supports high-throughput event streams.
   • Hadoop–Spark Integration
     • Spark works seamlessly within the Hadoop ecosystem.
       • Runs on YARN
         • Uses YARN for resource allocation.
         • Shares the cluster with other Hadoop tools.
         • Supports multi-tenant applications.
       • Reads/Writes from HDFS
         • Full compatibility with Hadoop storage.
         • Supports large-scale distributed files.
         • Enables high-speed parallel reads.
       • Replaces MapReduce
         • Faster in-memory execution.
         • Simplifies data pipelines.
         • Reduces latency for iterative jobs.

4. BIG DATA APPLICATIONS

   • Batch Processing
     • Large-scale offline data analysis.
       • Log Analysis
         • Processes massive log files.
         • Detects patterns and anomalies.
         • Supports system monitoring.
       • ETL Pipelines
         • Extract, Transform, Load workflows.
         • Data cleaning and preparation.
         • Automated batch data ingestion.
   • Real-Time Streaming
     • Continuous processing of incoming events.
       • IoT Data Streams
         • Sensor data processing.
         • Smart device communication.
         • Edge-to-cloud pipelines.
       • Clickstream Analysis
         • Tracks user behavior on websites.
         • Supports personalization.
         • Enables real-time marketing decisions.
   • Machine Learning
     • Data-driven predictive modeling.
       • Predictive Models
         • Forecasting future outcomes.
         • Uses regression and classification.
         • Supports risk scoring.
       • Recommendation Systems
         • Personalized suggestions.
         • Collaborative filtering models.
         • Real-time product recommendations.
   • Data Lakes
     • Centralized repositories for all data types.
       • Multi-format Storage
         • Stores structured, semi-structured, and unstructured data.
         • Supports CSV, JSON, Parquet, images, logs.
       • High Scalability
         • Expands horizontally by adding nodes.
         • Suitable for petabyte-scale analytics.

5. HADOOP AND THE NEW CHALLENGES OF GENERATIVE AI

   • The evolving role of Hadoop in the age of AI.
     • Data Volume & Scalability Needs
       • Massive training corpora
         • Text, images, audio, logs.
       • Horizontal scalability
         • Distributed storage for petabyte datasets.
       • Multi-format data handling
         • Unstructured and semi-structured data.
     • Speed & Low-Latency Requirements
       • Real-time analytics workloads
         • Stream ingestion for model updates.
       • Spark over MapReduce
         • In-memory execution suits AI needs.
     • Cloud & Lakehouse Evolution
       • Data Lakehouses
         • Delta Lake, Iceberg, Hudi on top of HDFS/S3.
       • Hybrid cloud deployments
         • On-prem + cloud storage integration.
     • Limitations of Traditional Hadoop
       • MapReduce is too slow
       • High operational complexity
       • Declining relevance

Bilder und Dateien der Mindmap