Boost Your Database Performance 10x with Oracle SecureFiles

Amit GaneshSenior DirectorOracle

Ravi RajamaniSenior Product ManagerOracle

Aris PrassinosDistinguished Member ofTechnical StaffMotorola

Agenda

• Enterprise Data• Current Approaches to managing data• SecureFiles: Changing the paradigm• High Performance• Advanced Features• Out-of-the-Box benefits with SecureFiles• Motorola – Real world experience• SecureFiles Interfaces• Summary

Enterprise Data

Audio, Video

Vector, Spatial

SQL Data

Documents

Image Data

XML Data

Emails, IM

Unstructured

65%

Structured20%

Sem

i-St

ruct

ured

15%

Yearly Data GrowthData Growth

Structured 15 - 20%

Semi & Unstructured 50 - 100%

* Gartner & IDC Estimates

• In a typical enterprise, Structured Data is ~20%• Semi & Unstructured Data represents the other 80%• Data growth is happening across the board!

Managing Information

• Organizations need to efficiently and securely manage all data

Structured Semi-Structured Unstructured

XML PDF

• Simplicity and performance of file systems makes it attractive to store file data in file systems, while keeping relational data in DB

• Mainstream DBs support ANSI-standard LOB for storing file data inside DB – performance is a concern for many users

Files Belong with Relational Data

• Many Enterprise applications manipulate both files and relational data• Rich user experience, compliance, business integration

• This split compromises security, robustness, and management• Disjoint security and auditing models• Changes cannot be made atomically• Backup and recovery are fragmented• Search across relational data and files is difficult• Space management is complicated• Separate interfaces and protocols• Application architecture more complex

Two data managers for one application is one too many!

Oracle SecureFilesConsolidated Secure Management of Data

• SecureFiles is a new database feature designed to break the performance barrier keeping file data out of databases

• Similar to LOBs but much faster, and with more capabilities• Transparent encryption (with Advanced Security Option)• Compression, deduplication (with Advanced Compression Option)• Preserves the security, reliability, and scalability of database• Superset of LOB interfaces allows easy migration from LOBs

• Enables consolidation of file data with associated relational data• Single security model• Single view of data• Single management of data

SecureFiles Innovations

• SecureFiles is a major rearchitecture of how the database handles unstructured (file) data• Not a new SQL data type, but new storage type for file data• Not an incremental improvement to LOBs

• Entirely new:• Disk format• Network protocol• Caching and locking• Redo and undo algorithms• Space and memory management• Cluster consistency algorithms

Designed for High Performance

• Write Gather Cache• Cache above the storage layer buffers data up to 4MB during writes

before flushing to disk• Allows for large contiguous space allocation for LOB data and

reduced write latency.• Intelligent Pre-fetching

• Improves read performance by pre-fetching LOB data from disk• Overlaps disk IO with network latency to improve throughput

• New Space Management routine• Automates new space allocation and “freed” space reclamation• Optimized chunk size reduces fragmentation

• No more High Water Mark contention as with old LOBs• Deletion and Reuse of entire LOBs not just individual chunks.

High PerformanceSecureFiles Vs. NFS

• SQL+ File test, single stream, single host• Using Secure Files is faster across the board

• 2x-3x faster for Queries, 6x for Inserts• Tests run using both SecureFiles and NFS/ext3 in metadata

journaling only (default for NFS)• Filesystem-like performance

File Read Performance File Write Performance

0.010.1

110

100

1 10 100 1000 10000

File Size (KB)

MB

/s

0.1

1

10

100

1 10 100 1000 10000

File Size (KB)

MB

/s

SecureFiles NFS

High PerformanceSecureFiles Vs. BasicFiles

050

100150200250300

0.01 0.1 1 10 100

MB

/s

Writes: New Space

File Size (MB)

050

100150200250300

0.01 0.1 1 10 100

MB

/s

Writes: Reused Space

File Size (MB)

SQL+ File Test: Concurrent Reads/Writes, OCI, 4 streams • Adding Files using New Disk Space: Up to 2x faster• Adding Files Reusing Space: Up to 22x faster• Reads up to 3x faster

050

100150200250300

0.01 0.1 1 10 100

MB

/s

Read Performance

File Size (MB)

SecureFiles LOBs

Advanced Features - Compression

• SecureFiles introduces Compression for unstructured data• Huge storage savings

• Industry standard compression algorithms• 2-3x compression for typical files (doc, pdf, xml)

• Automatically detects if SecureFile data is compressible• Skips compression for already compressed data• Auto-turn off compression when space savings are minimal or zero

• Server-side compression• Allows for random reads and writes to SecureFile data

• Two levels of compression provide different compression ratios• Compression Levels: MEDIUM (default), HIGH• Higher the degree of compression, higher the latency and CPU overhead

incurred

• Part of the Advanced Compression Option

Advanced Features - Deduplication

• Enables storage of a single physical image for duplicate data• Significantly reduces space consumption• Dramatically improves writes and copy operations

• No adverse impact on read operations• May actually improve read performance for cache data

• Specially useful for content management, email applications and data archival applications

• Part of the Advanced Compression Option

Secure hash

Advanced Features - Encryption

• Extends Transparent Data Encryption (TDE) functionality to SecureFile data• Data encrypted on disk• Key management completely transparent to applications

• Support for industry-standard encryption algorithms• 3DES168• AES192 (default)• AES256

• Unified security level for both file and relational data• Part of the Advanced Security Option

Out of the Box Benefits

• Easy conversion from LOBs to SecureFiles using Online Redefinition• Existing LOB applications benefit with no changes

• SecureFiles can be enabled only on new partitions• Requires no data migration• New data can benefit from high performance and other

advanced features of SecureFiles

• New installations• By setting db_securefiles= ALWAYS

• SecureFiles is fully integrated with• XML DB, Oracle Multimedia, Oracle Spatial, Oracle UCM

Motorola – Real World Experience

Aris PrassinosDistinguished Member of Technical Staff

Motorola Biometrics

• Trusted developer and integrator of Biometrics solutionsfor over 30 yearshttp://www.motorola.com/biometrics

Motorola PrintrakBiometric Identification Solution

• Capture, Processing, Search and Storage of multiple biometrics• Criminal and Civil applications

• Criminal investigation, background checking, border control, biometric documents, welfare disbursement, voter registration

Some Motorola Biometrics customers

• Motorola Supplies Mobile Fingerprint System for UEFA EURO 2008http://www.motorola.com/mediacenter/news/detail.jsp?globalObjectId=9822_9751_23&pageLocaleId=2026

• Motorola Plays Key Role in UK Biometric Visas Pilot Programmehttp://www.motorola.com/mediacenter/news/detail.jsp?globalObjectId=8778_8707_23

• Motorola Biometric Technology Wins Again at Super Bowl XLIhttp://www.motorola.com/mediacenter/news/detail.jsp?globalObjectId=7887_7816_23

• Motorola’s Biometric Technology Integral to the Registered Traveler Programhttp://www.motorola.com/mediacenter/news/detail.jsp?globalObjectId=7558_7486_23

… and of course the CSI crime labs

• Motorola Provides Miami Dade Police with Enhanced Biometric Identification Toolshttp://www.motorola.com/mediacenter/news/detail.jsp?globalObjectId=7299_7240_23

http://www.miamidade.gov/mdpd/BureausDivisions/bureau_CrimeSceneInvestigations.asp

A Motorola Printrak BIS system

• 12TB (projected 40TB)• 90% is LOBs (biometric images, XML metadata, files)• LOBs range from 5K to 5MB• 500 million LOBs (projected 1.5 billion)

• OLTP • All data current and may be read / updated / deleted

• 100 concurrent users• Java-based Service Oriented Architecture• 2-node Oracle 11g RAC

• 2 quad-core CPUs on each node

Images on the file system?

• Not an option for our customers!• Two data stores will diverge over time

• Non-transactional – Limited rollback capabilities• Almost certain data loss

• Incomplete recovery• Costly and complex high availability options

• Chain of custody• File system security and auditing limitations

• No onsite DBA or administrator • Manageability of too many moving parts• Space management and fragmentation

Images inside the database

• Storing images inside database as BasicFile LOBssince 2001 (starting with Oracle 8i)

• Over 80 Printrak BIS installations with BasicFiles • 500GB to 3TB in size• Resolved problems of using two data stores• Acceptable performance given the benefits of this approach• Market success and industry recognition

• Frost & Sullivan Award for Competitive Strategy Leadership (2004)

• Four Oracle Open World presentations since 2004

Metadata stored as XML

• Each customer requires a custom schema• Different applications (criminal, civil) • Different customer requirements for the same application

• From basic demographics to full criminal history• Schema to evolve over time with no application changes

• Storing metadata as schema-less XML• Tables same across installations (e.g. Person, Event etc)• Columns differ across installations and stored as XML• Relying heavily on XML indexing (Oracle Text, XMLIndex)• Underlying storage model for XML is CLOB

Files also inside the database

• Storing up to 200,000 regular files inside the database• Stored as LOBs in the XMLDB repository

• XMLDB still using BasicFiles in present release• Access via HTTP, Native Web Services, Windows Explorer• More reliable than NFS for concurrent access• RMAN Backup / Recovery - Disaster Recovery via DataGuard • Security and auditing

• SecureFiles file system (in upcoming release)• Filesystem API allows integration with existing applications • Addresses limitations of HTTP, WebDAV access• No third party packages such as davfs, wdfs for Linux access

Challenges with BasicFiles

• Low concurrency / RAC scalability due to contention• Hash partitioning helps (but with extra licensing costs)

• Data encryption customer requirements• BasicFiles not supported by TDE

• Significant disk space wastage (as much as 25%)• Inefficient CHUNKing due to widely varying image sizes• Large blocksize for performance at the expense of space

• Inefficient space reclamation • OLTP nature of application causes frequent deletes / updates

• Very competitive landscape on speed / price

Why are we excited with SecureFiles?

• SecureFiles allow high concurrency / RAC scalability• Hash partitioning still beneficial but not mandatory

• SecureFiles support Transparent Data Encryption• Very efficient space management

• Dynamic CHUNKing• Large blocksize no longer necessary for performance

• Very efficient space reclamation• Very pronounced effect in multi-terabyte OLTP databases

• No other RDBMS matches speed of SecureFiles• Observed 2x – 3x out of the box speedup over BasicFiles

for LOBs up to 5MB and full logging enabled

Tips on using SecureFiles

• Test with ARCHIVELOG• Behavior different • Performance drop larger than with standard datatypes

• Additional performance drop with FLASHBACK• Consider impact of FLASHBACK if using it for Data Guard

• Works out of the box without excessive tuning• Tune redo logs before anything else • Use fast disks for flash recovery area

Tips on using SecureFiles(cont.)

• Tune network parameters• Very important if accessing SecureFiles over slow WANs• Less noticeable effect on fast LANs

• sdu / tdu • 32767 a good value for sdu / tdu

• send_buf_size / recv_buf_size • depending on the bandwidth delay product of network

• OS kernel network parameters e.g. rmem, wmem• Very important to also set network parameters on the client

• If using thin JDBC use long version of the connect string

Tips on using SecureFiles(cont.)

• Carefully evaluate old tuning practices for BasicFiles• Not necessarily applicable or as important to SecureFiles

e.g. large blocksize, separate tablespaces, caching writes

• Performance benefits come from many places e.g. concurrency, pre-fetching, space reclamation• Test a realistic application scenario, not just single thread

performance• Efficient space reuse during insert-delete-reclaim processing

may result in a dramatic speedup over BasicFiles(10x – 20x as published in some benchmarks)

SecureFile Interfaces

• SecureFiles is 100% backward compatible with ANSI SQL 92 LOB interfaces

• SecureFiles can be accessed by both database clients and file system clients

• Database clients use standard LOB interfaces• JDBC, ODBC, OCI, OCCI, .NET, PL/SQL

• File system clients use the file system protocols• Implemented using the Oracle Secure File System• Available on Linux

Opening the floodgates to files

File systems

Database

JDBC, ODBC

Oracle Secure File System

• SFS is file system in the database, uses database for storage and brings all of database technology to file systems

• SFS implements the file system interfaces:• 2 methods (getpath, list) for a readonly file system• 5 methods for an a file system with read and write support• 15 methods for fully functional POSIX file system

• SFS interface is extensible for easily defining special purpose implementations (providers)• SFS can surface one or more DB tables as a filesystem or a single

table through multiple file systems• Example, a CheckImages table can have 2 filesystems on it:

• /CheckImages_by_customer/CustomerName/check.jpg• /CheckImages_by_date/2008/September/check.jpg

Linux Client for Oracle Secure File System

• Conceptually like NFS Client, but uses DB instead of NFS Server• User mode filesystem, no changes to Linux kernel

• Standards based POSIX compatible• Transparent to file system applications• Mount SFS at a Linux mount point at some path /mnt/sfs/…

• mount.lcos user@server:port/service /mnt/sfs• What can you do with it?

• Run E-Business Suite on SFS• MAA for mid-tier to host Log and Report directories on SFS

• Run Apache Web Server on SFS• Run virtual machines from SFS• Run YOUR custom application on the database

• Benefit from Security (encryption, SSL), HA (standby, hot backup), RAC (scale on commodity h/w), Compression/De-duplication,…

Secure File System performance

Secure File System Performance (4 way parallel)

0 100 200 300 400 500 600

1MB

10MB

100MB

File

Siz

e

MB/s

Secure File System WriteSecure File System Reads

Quiz

You accidentally deleted one file yesterday. How do you recover it?

How do you convert the file system into a compressed file system online?

How do you get a tamper-proof file system?

What technology can you use to Secure the file system?

The Best of Files and Databases

• SecureFiles have all the leading-edge file system capabilities • Deduplication, Encryption, Compression, Logging, Versioning

• SecureFiles have advanced database features not in file systems• Transactions, Read Consistency, Flashback• Readable Standby, Consistent Backup, Point in Time Recovery• Fine Grained Auditing, Label Security• Sliding Inserts• XML indexing, XML Queries, XPath• Real Application Clusters• Automatic Storage Management• Partitioning and ILM• Search across meta-data and file content

• Capabilities go far beyond any other database or file system• Having the best of both worlds removes the need to compromise

The Best of Files and Databases

• SecureFiles have all the leading-edge file system capabilities • Deduplication, Encryption, Compression, Logging, Versioning

• SecureFiles have advanced database features not in file systems• Transactions, Read Consistency, Flashback• Readable Standby, Consistent Backup, Point in Time Recovery• Fine Grained Auditing, Label Security• Sliding Inserts• XML indexing, XML Queries, XPath• Real Application Clusters• Automatic Storage Management• Partitioning and ILM• Search across meta-data and file content

• Capabilities go far beyond any other database or file system• Having the best of both worlds removes the need to compromise

Resources Available

• Oracle Technology Network (OTN)• http://www.oracle.com/technology/products /database /

securefiles/index.html• Technical White paper• Performance White paper• Data Sheet• Internet Seminar• SecureFiles Test Kit

• SecureFiles Tutorial• http://www.oracle.com/technology/obe/11gr1_db/datamgmt/se

curefile/securefile.htm

Sessions From Oracle Development

• S298741 - Oracle Database 11g: Next-Generation Performance and Scalability, 11:30 - 12:30 pm,Moscone South 103

• S298792 - Real-World Database Performance Techniques and Methods, 12:00 - 1:00 pm, Moscone South 104

• S298786 – Current Trends in Real-World Database Performance, 1:00 - 2:00 pm, Moscone South 103

Monday, Sep 22

Wednesday, Sep 24

Thursday, Sep 25

• S298759 – Oracle Advanced Compression: Throw Away Half of Your Disks, and Run Your Database Faster, 9:00 - 10:00 am, Moscone South 103

• S298756 - Boost Your Database Performance 10x with Oracle SecureFiles, 11:30 - 12:30 pm,Moscone South 104

• S298785 – Real-World Database Performance Roundtable, 1:30 - 2:30 pm, Moscone South 104

Tuesday, Sep 23

• S298757 – Time-Travel Through Your Data with Oracle Total Recall, 11:30 - 12:30 pm, Moscone South 305

• S298754 – Using Oracle Database in the Cloud , 1:00 - 2:00 pm, Moscone South 305

• S298765 – Oracle Database Performance on Flash Drives: Insights from Oracle Development , 9:00 -10:00 am, Moscone South 104

Labs/Demos From Oracle Development

Total Recall

Oracle DEMOgrounds, Moscone WestMonday, Sep 22 – Thursday, Sep 25

Hands-on LabsMarriott Golden Gate A3

Monday, Sep 22

2:30 – 5:00 pm

Advanced Compression

Tuesday, Sep 23

5:00 – 6:00 pm

Total Recall

Thursday, Sep 25

1:30 – 4:00 pm

Advanced Compression

Q U E S T I O N SA N S W E R S

For More Information

search.oracle.com

ororacle.com

<Insert Picture Here>

Appendix

SecureFiles Best Practices

• CHUNK parameter no longer needs to be specified• SecureFiles automatically optimizes writes to use maximum

sized contiguous allocations on disk – reduces fragmentation

• Write Gather Cache• Buffers writes across calls for large, contiguous I/O• Recommend inserting multiple rows between commits,

especially for applications inserting small/medium sized LOBs

• New logging level, FILESYSTEM_LIKE_LOGGING provides logging similar to most journaled file systems• Only SecureFiles metadata is logged, not SecureFiles data• Significantly reduces amount of data written to Redo logs

SecureFiles Best Practices

• Recommend dedicated tablespaces and devices for SecureFiles segments• Allows for large, contiguous allocations for writes• Efficient pre-fetching for reads

• Database parameters• SDU/TDU Size – set to maximum possible value for more

efficient streaming of SecureFiles data• Recv_buf_size/Send_buf_size – Tune to higher value to have

significant impact on network throughput• Kernel parameters

• Increasing network buffer sizes greatly increases SecureFiles throughput