The Architecture of Inspur Cloud Storage 2013.7
Contents 1 2 What is the problem What is our solution The Architecture & Key Design Cloud Computing with Cloud Storage 3 Conclusion
Contents 1 2 What is the problem What is our solution The Architecture & Key Design Cloud Computing with Cloud Storage 3 Conclusion
What is in the Cloud web Massive Data create/obtain table electronic document destroy/delete Index/identify printout equipment paper safe/protect search Images, video store/save share/collaborate email Business documents faxes instant messaging data base
What the Cloud Storage must support Massive Data GB, TB, PB, EB All Data Format Structured Data Semi-Structured Data Unstructured Data Multi Access Mode Block Device S3 Swift NFS HDFS Data Safety Performance Advanced Data Management Even More
Contents 1 2 What is the problem What is our solution The Architecture & Key Design Cloud Computing with Cloud Storage 3 Conclusion
Our Solution Inspur Cloud Storage VM FS DB Browser Map Reduce App iscsi, FCP, iser Inspur WOS NAS Block Interface S3 & Web API Inspur WOS Cloud REST Interface WOS Native Object Inspur Interface Native Object Interface C++ Python Java HTTP
The Architecture Unified Storage Pool + Multi-Protocol Access Unified Storage Pool: Integrate hardware resource Multi-Protocol Access: Block Device, REST(S3, Swift), FS, etc MPA Block Device REST FS C++/Java/ Python USP Monitors Storage
Key Components: Unified Storage Pool USP provides service by object Each object has a unique name (object id) Object is located by a Pseudo-Random Consistent Hashing No relations between objects and Fully Scale-Out object object storage devices 9
Key Components: Unified Storage Pool USP distribute data to different rack, node and disk. The whole cluster is described by a topological diagram which we called NodeMap, while NodeMap is used for object location algorithm. We treat hardware resource changes including expansion, reduce and breakdown as change of NodeMap. root racks 1 2 3 nodes
Multi-Protocol Access Multi-Protocol Access Layer accepts client requests with responsibility for common functions such as Balance, Line Redundancy and Consistency. Different interfaces handle different request format with the same USP Access Model FS Client SAN Client Web Client Multi-Protocol Access Layer Balance Redundancy Consistency FS Interface Block Interface REST Interface Meta Data File Data Volume Manage HTTP Server USP Access Model USP Interface
Contents 1 2 What is the problem What is our solution The Architecture & Key Design Cloud Computing with Cloud Storage 3 Conclusion
Cloud Computing System Based on Cloud Storage Two Typical Modes In Cloud Computing Virtual Machine(VM Ware, KVM, Xen) Big Data Analyze(MapReduce) Inspur WOS NAS Block Interface Inspur WOS Cloud HDFS Interface
Block Device Access Block Device Access Model combines object collection into a standard Linux block device. Thin-provisioning, Clone, Isolation and Access Control is supported MPA Block Device Thinprovisioning Clone Isolation USP Monitors Storage
Block Device Access The Key of Block Device Access Model is Sparse Data Map Algorithm BDAM is implemented as a Linux driver(icsb) and the virtual block device meets the vblock interface Sparse Data Map Algorithm maps the vblock request(offset, length) to Object request(object id, offset, length) Block Device /dev/icsb1 Object Collection a 1 a 2 a 3 a 4 a 5 a 6 a 7
Block Device Access Applications VM KVM Xen FS DB vblock (offset, length, data) icsb /dev/icsb1 /dev/icsb2 map combine map combine objects o 1 o 2 o 3 o 4 o 5 o 6 o 7 o 8 USP
Map Reduce Task 1 data1 data2 data3 Task 2 data4 data5 data6 object 1 object 2 object 3 object 4 object 5 object 6
Contents 1 2 What is the problem What is our solution The Architecture Key Components 3 Conclusion
Conclusion Key Points Unified Storage Pool + Multi-Protocol Access USP: Object is Located by Pseudo-Random Consistent Hash MPA: Different interfaces with the same USP Access Model Cloud Computing: Block Device Access & HDFS API Exabyte Data or Even More TB/s Bandwidth Millions of Concurrent Request Capability
Thank you Inspur group 7/24/2013