European Archival Records and Knowledge Preservation Database Archiving in the E-ARK Project

Transcription

1 European Archival Records and Knowledge Preservation Database Archiving in the E-ARK Project Janet Delve, University of Portsmouth Kuldar Aas, National Archives of Estonia Rainer Schmidt, Austrian Institute Technology DLM Forum, 13 November 2014

2 THE E-ARK PROJECT IS CO-FUNDED BY THE EUROPEAN COMMISSION UNDER THE ICT-PSP PROGRAMME

3 Outline E-ARK objectives Current practices and needs Transactional (OLTP) vs Analytical (OLAP / Data Warehousing) techniques Database archiving in E-ARK

4 E-ARK Facts and Figures EU CIP PCP ICT Programme Objective 2.5: earchiving services Pilot B The pilot should share information on integration, operation and interoperability issues throughout the EU in order to facilitate the creation and maintenance of a European archiving infrastructure for government and public services thus promoting the re-use of archival data. 36 months: February 2014 January M Budget, 3 M funded by EC 16 Partners for all deliverables for all software 4

5 E-ARK Objectives Reduce the cost of transfer, preservation and access to digital information by Standardising how agencies export and send information to digital archives Providing open formats for the long-term preservation of various content Exploring needs for accessing archives and providing novel interfaces for these Long-term vision: Improve semantic and technical interoperability to a level which allows any system developer to deliver out-of-the-box archiving functionality

6 Pre-Ingest E-ARK SIP SIP Creation Tools Archival records Content and Records Management Systems Ingest and Preservation SIP AIP Conversion Digital preservation systems Scalable Computation E-ARK AIP AIP - DIP Conversion CMIS Interface Data Mining Interface E-ARK DIP Access Archival Search, Access and Display Tools Content and Records Management Systems Data Mining Showcase

7 Current database archiving practice Snapshot policy Ingest: Transform the original relational structure into open formats Formats: SIARD, ADDML, DBML Access: Users need to find the appropriate snapshot(s), load these into a current DBMS and use predefined queries or build their own ones DB Snapshot Table 1 PK Row 1.1 Row 1.2 Row 1.3 Table 2 PK Row 2.1 Row 2.2 Row 2.3 Table 3 PK Row 3.1 Row 3.2 Row 3.3 Codes Code 1 Code 2 Code 3 Ingest Access

8 Problems Finding the appropriate snapshot Most users search for data about something Which car had the plate number 111YYY in January 15th 2000 Current practice allows to search for the database snapshot which includes data about something Which database includes information about cars in January 15th 2000 Scope of the snapshot The scope of data and time period covered in a single snapshot usually do not meet the needs of the user Required technical knowledge Relational structures are often highly optimised and hard to grasp Most users do not have the knowledge to build accurate queries for specific access needs The only way is to use pre-defined queries which have been archived along with the data

9 In the ideal world users do not need to search for databases but data! Semantic reuse Topic based reuse Big data analysis

10 HOW TO DO IT?

11 Transactional Processing (OLTP)

12 Online Analytical Processing (OLAP) OLAP

13 Data warehousing Updates only from a DB Snapshots (Useful for DB archiving) Can be denormalised Star schema dimensional model

14 Star Schema

15 Database archiving in E-ARK Overview of E-Ark Concepts Archiving of databases in different layers: primary format, semantic representation, representation for analytical processing. Data intensive technology for AIP storage and processing. Hadoop, HDFS, HBase, Lily, SolR. Support for database transformation and analysis such as denormalization, aggregation, indexing. Levels of DIP format and display Access to archived records, based on OLAP queries and reports, as dynamically reconstructed RDBs.

16 Extract Transform Load Goal: Integrate data from multiple applications into a database / warehouse. Extracting data from source systems like RDBMS and flat files. Transform: derive, extract, aggregate data. Load data into target: Overwrite cumulative information or add new data Important pre-processing step for data mining/analytics. involves data cleaning and data integration Result: Structured data, random access based on data based indexes (e.g. RDBMS, NOSQL). E-Ark Approach: Automated transformation of archived databases into snowflake schema representation(s). Denormalized, connected fact and dimension tables

17 Indexing - Full Text Search Searching documents based on full text distinguished from searches based on metadata Returns (ranked) list of document IDs Involved Information Retrieval methods Building an inverted index Scoring and weighting Results Text classification Evaluation Approach in E-Ark Denormalization / star schema transformation and ingestion into Apache HBase. Repository and faceted search on records based on NGDATA s Lily repository and Apache SolR.

18 Online analytical processing (OLAP) OLAP Database / Data Warehouse Aggregated, historical data, low transaction rate Resource-intensive and complex queries Analyse multi-dimensional data in a read efficient manner (Web analytics, sales) View metrics by combination of dimensions Time vs. Space: Pre-aggregates data to build cube Dynamically analyze data from multiple perspectives roll-up, drill-down, and slicing and dicing Approach in E-Ark Data analytics based on pre-processed database representation arranged along dimensions. Data loaded and queried through Apache HBase. Access (DIPs) supported by additional use of OLAP tools.

19 Data Mining Identify correlations and patterns in existing data Used by statisticians, database and business communities Data Analysis Techniques Regression: predict continuous valued output (e.g. price) Classification: discrete valued output (e.g. char. recognition) Segmentation: Separates data into interesting groups Based on mathematical methods pattern matching, machine learning, numerical analysis E-Ark data mining mainly based on text mining. Using data structure of the repository or search index Scalable through MapReduce / Apache Mahout. Goal: Clustering, Labeling, Anomaly Detection

20 Proposed Architecture EARK-AIP Data Management Application ESS Arch Preservation Platform Data Mining Showcase T6.4 D6.3 Data Connector API CRUD API Query API Data Mining API AIP Storage T6.2 MS10 Data Management Integration T6.1 MS06, D6.2 Query and Indexing T6.3 MS04, D6.1 Scalable Computation Staging Area Lily, Hadoop, HBase, HDFS Re-use and Data Mining T6.4 Archive Storage (WORM)

21 Tasks and Components Archival Storage Store APIs on HDFS using ESS Preservation Platform Bulk-load, permanent and replicated storage Data Integration Extract data from archival information package. ETL data into Lily/HBase, keep AIP in HDFS (don t touch) Query and Indexing Metadata on AIP level stored in HBase for basic retrieval Faceted search based on Apache SolR Data Mining and Analytics Load OLAP structure from package Data sets stored on record level into HBase Query for facts based on different dimension and levels.

22