AI 용어집
인공지능 완전 사전
GFS (Google File System)
Distributed file system architecture developed by Google, optimized for Big Data workloads with sequential data access and automatic replication management.
NameNode
Master node in HDFS that maintains the file system metadata, including directory structure, block locations, and cluster topology.
DataNode
Worker node in HDFS responsible for the physical storage of data blocks, their replication, and communication with the NameNode for status reports.
Block Replication
Fault tolerance mechanism where each data block is stored in multiple copies on different DataNodes to ensure data availability in case of hardware failure.
Rack Awareness
Algorithmic optimization that places data replicas on different physical racks to maximize fault tolerance in case of an entire rack failure.
HDFS Federation
Scalable HDFS architecture allowing multiple independent NameNamespaces, each managing a portion of the namespace to improve scalability and isolation.
Checkpoint Node
Specialized node in HDFS that periodically creates file system checkpoints by merging the modification log with the file system image to optimize restart.
Erasure Coding
Data protection technique more efficient than replication, dividing data into fragments and encoding parity fragments to reduce storage space while maintaining fault tolerance.
FsImage
Persistent snapshot of HDFS filesystem metadata, containing the complete directory structure and file attributes at the time of its creation.
Edit Log
Transactional log in HDFS that sequentially records all filesystem metadata modifications to ensure consistency and enable crash recovery.
Safe Mode
Read-only state of HDFS during which the NameNode verifies the minimum replication of blocks before accepting modifications, ensuring data consistency at startup.
Distributed Namespace
Architecture where metadata management is distributed across multiple servers to eliminate bottlenecks and enable horizontal scalability of the filesystem.
Block Placement Policy
Algorithm determining the optimal placement of block replicas in the cluster by considering network topology, node load, and fault tolerance requirements.
Journal Node
Component in HDFS High Availability that persistently stores the active NameNode's modification logs to enable seamless failover to the standby NameNode.
Data Balancer
Tool that redistributes blocks between DataNodes to balance disk space usage and network load in the cluster, optimizing overall performance.
Heartbeat Mechanism
Periodic signal sent by DataNodes to the NameNode to confirm their operational status, enabling rapid failure detection and triggering replication.
Namespace Quota
Configured limit on the number of files and directories that a user or group can create in HDFS to control system resource usage.