Transcript Multimedia Storage Management

Multimedia Storage Management
MS
I
Sunil Prabhakar
Assistant Professor
Department of Computer Sciences
Purdue University
Researchers
MS
I
W. Aref (CS)
 A. Elmagarmid (CS)
 A. Ghafoor (ECE)
 R. L. Kashyap (ECE)
 M. Khan (ECE)
 S. Prabhakar (CS)

2
On-Going Research Projects for
Multimedia Storage Management
MS
I



Efficient image storage
• Disk placement schemes
Tertiary storage management
• I/O scheduling and reliability for robotic
libraries
Multimedia document management
• Document placement techniques
• I/O management
Goal: improved performance
3
Image Placement for
Multiresolution Browsing
MS
I
Images decomposed into multiple resolutions
using wavelets
4
Image Placement on Disks
MS
I
Content-based retrieval using browsing
graphs or quantitative measures
 Multidimensional data declustering for
similarity queries
 Designed and evaluated (through
simulation and implementation) several
efficient disk placement schemes

5
Tertiary Storage Management
MS
I
Scheduling and reliability for tertiary
storage are critical issues. These issues
arise due to high latency and rapid tape
wear-out
 Most scheduling work has focussed on
single medium scheduling
 Work on reliability has been limited to
mirroring or RAIT

6
Research Results for Tertiary
Storage Management
MS
I

Scheduling
• Developed scheduling algorithms
– Optimal for single drive libraries
– Near-optimal for multiple drive libraries
• Significant improvement over FCFS

Tape reliability
• Tape Group Parity Protection
– Lower storage overhead than mirroring
– Better performance than RAIT
7
Challenges for Multimedia
Document Placement
MS
I


Mixed media storage (audio, video, text,
images)
Mixed media data placement techniques
• Statistical techniques using access patterns
• Document striping across multiple disks

Document clustering and allocation
techniques for parallel disk arrays and
clusters of networked servers
8
Multimedia Browsing Graph
MS
I
Abomasum
Omasum
Reticulum
Protozoa
Esophagus
Text
Rumen
Audio
Text
Image
Stomach
Text
Image
Digestive System
Small Intestine
Salivary Glands
Large Intestine
Text
Video
Audio
Image
9
I/O Management Issues
MS
I
I/O channels between various levels of
the storage hierarchy are limited
resources
 To ensure QoS at run-time intelligent
I/O channel allocation schemes are
needed.
 Document profiles can be utilized for
dynamic I/O channel allocation

10
Document Bandwidth Profile
MS
I
Bandwidth
Exit
MICROBIAL
FERMENTATION:
Omasum
Abomasum
Microbial fermentation is an
anaerobic process
Reticulum
Esophagus
Protozoa
Rumen
Stomach
Digestive System
Small Intestine
Salivary Glands
Large Intestine
Time
Microbes live in a symbiotic
existence with the host (e.g.
Cow)
Microbes do not have
complete TCA cycle, volatile
fatty acids (VFA) are
byproducts of their anaerobic
Multimedia Browsing Graph
11
Related Work
MS
I  Image storage
• Chabot (UC Berkeley)
• QBIC (IBM)

Multimedia storage
•
•
•
•
•
Fellini (Lucent)
Mitra (USC)
IBM (Phillip Yu et al)
NASD
Streaming RAID (Tobagi et al)
12
Planned Multi-Level Storage
Architecture Environments
MS
I
Parallel Disk Array Architecture
Clustered Server Architecture
Cluster
RAID
Multimedia
Document
Profiles &
Traces
Tertiary Storage
Multimedia
Document
Profiles &
Traces
Tertiary Storage
13
Experiments for Multi-Level Storage
Architecture Environments
MS
I
Objective: Comparison between the two
environments in terms of QoS
guarantees
• Real-time scheduling
• Multimedia data placement schemes at
different levels of the storage hierarchy
• I/O channel management
• Reliability
• Scalability
14