Hard Disk Drive Evolution & Data Storage Systems Program

Information Storage Industry Consortium

Giora J. Tarnopolsky TarnoTek - www.tarnotek.com Storage on the Lunatic Fringe SC2003 - Phoenix - AZ 19 November 2003

z Who/What/When INSIC

z Exhilarating growth of volumetric density

zDrive economics

zIs it likely to continue?

z 1 Terabyte drive likely

zINSIC’s research programs

z EHDR

z HAMR

z INSIC & Data Storage Systems Research

G. Tarnopolsky /November 2003/ 2 © 2003 INSIC SC 2003 Information Storage Industry FLORIDA INTERNATIONAL UNIVERSITY DATA STORAGE INSTITUTE (DSI) LOS ALAMOS NATIONAL LAB Consortium CENTRAL LANCASHIRE COLORADO STATE JOHNS HOPKINS NORTHEASTERN IBM UC SAN DIEGO ECD* MANCHESTER UC BERKELEY IDC* z collaborative research consortium OHIO STATE SONY COLORADO PLYMOUTH MAXELL MISSOURI IMATION for the worldwide information NEBRASKA APRILIS* VIRGINIA ALABAMA QUANTUM HARVARD SAMSUNG storage industry ALBERTA ARIZONA CERTANCE ILLINOIS MAGNECOMP* z established 1991 MIT STORAGETEK ISIC DOWA MINING* NIST MEMS OPTICAL O IDEMA Conduct Joint Research on High Risk Pre-competitive IDAHO AGERE SYSTEMS Storage Technologies MITRE WESTERN DIGITAL PURDUE TORAY INDUSTRIES O Develop Technology Roadmaps STANFORD HEWLETT PACKARD MINNESOTA O Maximize Value of University Research VANDERBILT VEECO INSTRUMENTS SANTA CLARA ADVANCED RESEARCH O Obtain Government Funding GEORGIA TECH SEAGATE TECHNOLOGY ARIZONA STATE O Speak for the Industry NORTHWESTERN EUXINE TECHNOLOGIES CARNEGIE MELLON HUTCHINSON TECHNOLOGY ARGONNE NAT’L LAB HITACHI GLOBAL STORAGE TECHNOLOGIES WASHINGTON UNIVERSITY LAWRENCE BERKELEY NAT’L LAB LAWRENCE LIVERMORE NAT’L LAB * Limited Member NATIONAL UNIVERSITY OF SINGAPORE

G. Tarnopolsky /November 2003/ 3 © 2003 INSIC SC 2003 The Future of Hard Disk Drive Technology Lab Demos: Possible HDD Areal Density Progression 100000.0 Laboratory Demonstrations ne li 10000.0 te ra AG C % ) 2 1000.0 50 1 Terabit per inch2 goal

perpendicular recording 100.0 highest in products

Areal Density (Gb/ in Areal Density (Gb/ 70 heat-assisted mag recording now 10.0 patterned media ?

self-organized arrays ?

1.0 Jan-90 Jan-93 Jan-96 Jan-99 Jan-02 Jan-05 Jan-08 Jan-11 Jan-14 Jan-17 Date

10000.0

? )

2 1000.0

ne G li CA 100.0 30% Areal Density (Gb/ in Areal Density (Gb/

10.0

1.0 Jan-90 Jan-93 Jan-96 Jan-99 Jan-02 Jan-05 Jan-08 Jan-11 Jan-14 Jan-17 Date

Cost per Gigabyte, 95 mm Desktop ≤ 7,200 rpm 100.00 HGST Maxtor Seagate 10.00 WDC

1.00 - 44%/ year $/GB

0.10

0.01 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09

Price of HDD, 95 mm Desktop ≤ 7,200 rpm

1,000

$/Drive 100

HGST Maxtor Seagate WDC

HDD Capacity vs. Time, 95 mm Desktop ≤ 7,200 rpm 1000

100

All HDD Early HDD 10 HDD Capacity (GB)HDD Capacity

1 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09

HDD Capacity vs. Time, 95 mm Desktop ≤ 7,200 rpm 1000 INSIC 1 Tb/inch2 ? demo goal y = 6E-28e0.0018x 2 100 R = 0.9626

All HDD Early HDD 10 HDD Capacity (GB)HDD Capacity

1 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09

can’t z INSIC’s Extreme High write Density Recording program strives for concurrent high unstable SNR, permanency of the goal recorded bit, and ability to record (EHDR)

z INSIC’s Heat Assisted

signal-to- Magnetic Recording noise ratio program uses heat to recordability th achieve recordability e rm (HAMR) a l s z Patterned media ta b i li z Tilted perpendicular media ty low SNR z Self-organized media

Parameters System Areal density 1000 Gb/in2 Capacity 8 TB Linear density 2500 kbpi Drives 8 Track density 400 ktpi Physical size Rotation 15,000 rpm Width 16 in Disk 33 mm Height 2 in Capacity 125 GB/platter Depth 6 in Internal transfer rate 320 Mbyte/s Drive z Adapted from R. Wood, Capacity 1 Terabyte "The feasibility of magnetic Platters 8 recording at 1 Terabit per Heads 16 square inch," IEEE Trans. Width 2 (50) in (mm) Magn. 36, No. 1, 36- Height 1.4 (36) in (mm) 42(2000) Length 4 (100) in (mm)

zz “Since“Since thethe evolutionevolution ofof SCSISCSI inin thethe mid-1980s,mid-1980s, thethe functionalfunctional definitiondefinition ofof aa diskdisk (or(or aa tape)tape) hashas beenbeen essentiallyessentially constantconstant …… thethe basicbasic modelmodel ofof aa singlesingle vectorvector ofof numberednumbered blocksblocks hashas remainedremained thethe same.”same.” zz “Researchers“Researchers todaytoday areare questioningquestioning whetherwhether thethe tried-and-truetried-and-true diskdisk (or(or tape)tape) functionalfunctional modelmodel isis thethe mostmost effectiveeffective wayway toto provideprovide networknetwork storagestorage services.”services.” (text(text inin italicsitalics addedadded byby GT)GT)

z Implementation & management of Terabyte drives & systems

Work that:

z is not now subject of industrial competition towards commercialization of a product

z addresses goals that are beyond the immediate industrial roadmaps

z may tackle high-risk/high-reward speculative ideas

z enhances the research return-on-investment to the benefit of all partners and of the global data storage systems endeavor

z Advances whose implementation would require industrial consensus among parties that otherwise may compete in the marketplace

z e.g. storage-device OS

z Pre-competitive research is, essentially, research “co-opetition.”

z Are there pre-competitive research opportunities in data storage systems work?

Brainstorm the Future

z Data storage systems, software, and storage devices’ companies

z Academics & academic centers

z Entities vested in the storage enterprise: government, sponsored research efforts with DSS aspects

z Current sponsors: INSIC, CMRR UCSD, DTC/DISC U Minn, INSIC member companies 2 z In progress: CAL IT , others

z Research topics will emerge from industrial & academic consensus

z Gratefully acknowledge discussions with many researchers in creating a roster of candidate topics.

z David B. Anderson / Seagate z Ethan Miller / UC Santa Cruz

z Walter Burkhard / UC San Diego z Reagan Moore / SDSC/UCSD

z Randal Burns / John Hopkins University z Richard New / Hitachi GST

z David Du / DTC, U Minn z David Patterson / UC Berkeley

z Greg Ganger / CMU z Michael O. Rabin / Harvard University and Hebrew

z Garth Gibson / CMU & Panasas University

z Gordon F. Hughes / UC San Diego z Erik Riedel / Seagate Research

z Joseph F. JaJa / U. Maryland z Tom Ruwart / DTC/DISC U Minn

z Darrell Long / UC Santa Cruz z Paul Siegel / UC San Diego z Ed Skalko / Seagate Research z Patrick Mantey / UC Santa Cruz Thanks! z Paul Massiglia / Veritas Software z Barry Schechtman / INSIC z Boleslaw K. Szymanski / Rensselaer Polytechnic Institute

networked, DigitalDigital SemanticSemantic permanencypermanency distributed continuitycontinuity PerformancePerformance PerformancePerformance && reliabilityreliability DigitalDigital OSD, object- && reliabilityreliability permanencypermanency oriented storage hardware-ish device software-ish

Application-Application- Non-volatile,Non-volatile, lowlow awareaware storagestorage latencylatency memorymemory inin device device limited Multi-dimensional space and locale projections

z There is no assured scheme for perpetual content preservation

z Semantic continuity: make computer languages evolve like natural languages, assure comprehension

z Storage management independent of the medium and of the content itself

z Systems hold “eternal” data in devices bearing a ~ three-year warranty§,*

z Digital assets have undergone migrations to devices of higher performance & volumetric density.

No more. Safe harbor: §) H/W MTBF ≥ 30 yr. *) Tape cartridges guaranteed “for life”

G. Tarnopolsky /November 2003/ 21 © 2003 INSIC SC 2003 Large, non-volatile storage stratum z Multi-GB, ≤ µs-latency, non-volatile memory z Flash, M-RAM, or MEMS z Non-volatile stratum would be “virtual disk” such as disk is “virtual tape” to tape system z Various application archetypes could have assigned non-volatile storage streams - until objects of associated types are transferred to specific areas on the media

z Intelligent space allocation, self-

defragmentation 500 GBMag 12 GBNV TT

z OSDs take the storage-device-specific component of the file system into the storage device itself

z Ability of device to manage its own capacity

z Ability of device to export file-like objects to their clients

z Where in the storage hierarchy is the OSD concept to be applied? Disk remains byte- bucket but RAID is intelligent storage?

z Digital storage has irreversibly changed the transaction of human activity

z Change has occurred rapidly, and will continue

z Change will be driven by global industry & academic efforts

z Plethora of unresolved issues

z INSIC recognizes significant opportunities for pre-competitive, industry/academia cooperative data-storage-systems research, and will promote it.

INSIC PROGRAMS

Paul D. Frank Executive Director