Hadoop World 2011: HDFS Federation - Suresh Srinivas, Hortonworks
5 likes1,117 views
The document discusses HDFS architecture, focusing on its limitations such as scalability, isolation, and tight coupling between namespace and block management. It introduces HDFS federation, which allows multiple independent namenodes and namespace volumes, enhancing scalability and enabling more flexible storage solutions. Next steps include further separation of namespace and block management layers and improving overall efficiency in managing namespaces.
Hadoop World 2011: HDFS Federation - Suresh Srinivas, Hortonworks
- 2. Agenda • HDFS Background • Current Limitations • Federation Architecture • Federation Details • Next Steps
- 3. HDFS Architecture Two main layers • Namespace Block Storage Namespace › Consists of dirs, files and blocks Namenode NS › Supports create, delete, modify and list files or dirs operations Block Management • Block Storage Datanode … Datanode › Block Management Storage • Datanode cluster membership • Supports create/delete/modify/get block location operations • Manages replication and replica placement › Storage - provides read and write access to blocks 3
- 4. HDFS Architecture… Implemented as Block Storage Namespace Namenode • Single Namespace Volume NS › Namespace Volume = Namespace + Block Management Blocks • Single namenode with a namespace Datanode … Datanode › Entire namespace is in memory Storage › Provides Block Management • Datanodes store block replicas › Block files stored on local file system 4
- 5. Limitation - Scalability Scalability • Storage scales horizontally - namespace doesn’t • Limited number of files, dirs and blocks › 250 million files and blocks at 64GB Namenode heap size • Still a very large cluster • Facebook clusters are sized at ~70 PB storage Performance • File system operations throughput limited by a single node › 120K read ops/sec and 6000 write ops/sec • Easily scalable to 20K write ops/sec by code improvements 5
- 6. Limitation - Isolation Poor Isolation • All the tenants share a single namespace › Separate volume for tenants is desirable • Lacks separate namespace for different application categories or application requirements › Experimental apps can affect production apps › Example - HBase could use its own namespace 6
- 7. Limitation – Tight Coupling Namespace and Block Management are distinct services • Tightly coupled due to co-location • Scaling block management independent of namespace is simpler • Simplifies Namespace and scaling it Block Storage could be a generic service • Namespace is one of the applications to use the service • Other services can be built directly on Block Storage › HBase › Foreign namespaces 7
- 8. Isolation is a problem for even small clusters 8
- 9. HDFS Federation Namespace NN-1 NN-k NN-n Foreig NS1 NS k n NS n ... ... Pool 1 Pool k Pool n Block Storage Block Pools Datanode 1 Datanode 2 Datanode m ... ... ... Common Storage • Multiple independent Namenodes and Namespace Volumes in a cluster › Namespace Volume = Namespace + Block Pool • Block Storage as generic storage service › Set of blocks for a Namespace Volume is called a Block Pool › DNs store blocks for all the Namespace Volumes – no partitioning
- 10. Key Ideas & Benefits • Distributed Namespace: Partitioned across namenodes Alternate NN › Simple and Robust due to independent masters Implementation HBase HDFS • Each master serves a namespace volume Namespace MR tmp • Preserves namenode stability – little namenode code change › Scalability – 6K nodes, 100K tasks, 200PB and 1 billion files • Block Pools enable generic storage service Storage Service › Enables Namespace Volumes to be independent of each other › Fuels innovation and Rapid development • New implementations of file systems and Applications on top of block storage possible • New block pool categories – tmp storage, distributed cache, small object storage • In future, move Block Management out of namenode to separate set of nodes › Simplifies namespace/application implementation • Distributed namenode becomes significantly simpler
- 11. HDFS Federation Details • Simple design › Little change to the Namenode, most changes in Datanode, Config and Tools › Core development in 4 months › Namespace and Block Management remain in Namenode • Block Management could be moved out of namenode in the future • Little impact on existing deployments › Single namenode configuration runs as is • Datanodes provide storage services for all the namenodes › Register with all the namenodes › Send periodic heartbeats and block reports to all the namenodes › Send block received/deleted for a block pool to corresponding namenode 11
- 12. HDFS Federation Details… • Cluster Web UI for better manageability › Provides cluster summary › Namenode list and summary of namenode status › Decommissioning status • Tools › Decommissioning works with multiple namespace › Balancer works with multiple namespaces • Both Datanode storage or Block Pool storage can be balanced • Namenode can be added/deleted in Federated cluster › No need to restart the cluster • Single configuration for all the nodes in the cluster
- 13. Managing Namespaces • Federation has multiple namespaces – Don’t Client-side you need a single global namespace? / mount-table › Key is to share the data and the names used to access the data • A global namespace is one way to do that data project home tmp • Client-side mount table is another way to share. › Shared mount-table => “global” shared view › Personalized mount-table => per-application NS4 view • Share the data that matter by mounting it • Client-side implementation of mount tables NS1 NS2 NS3 › No single point of failure › No hotspot for root and top level directories
- 14. Next Steps • Complete separation of namespace and block management layers › Block storage as generic service • Partial namespace in memory for further scalability • Move partial namespace from one namenode to another › Namespace operation - no data copy
- 15. Next Steps… • Namenode as a container for namespaces › Lots of small namespaces • Chosen per user/tenant/data feed • Mount tables for unified namespace • Can be managed by a central volume server … Namenodes › Move namespace from one container to another for balancing Datanode … Datanode • Combined with partial namespace › Choose number of namenodes to match • Sum of (Namespace working set) • Sum of (Namespace throughput) 15
- 16. Thank You More information 1. HDFS-1052 – HDFS Scalability with multiple namenodes 2. An Introduction to HDFS Federation – https://hortonworks.com/an-introduction-to-hdfs-federation/