Overview of DLSS Workshop

1
Overview of DLSS Workshop

• Don Petravick
• FNAL
• Office of Science Data Management Workshop

2
DLSS

• What was it?
• DLSS Disk in Large Scale Storage Systems
• Feb 19-20
• Hosted by NASA Goddard
• http//www.westoverconferences.com/dvt/dvt_agenda.
  htm
• Agenda
• Government Requirements, Cost, Performance,
  Reliability, File Systems, Security, Legal
  Requirements.
• Ill present to my own schema.

3
Government Requirements

• Evident difference between defense type systems
  and scientific systems.
• Defense type systems
• stringent and increasing security requirements,
• kernel level file protections, etc.
• Science
• more able to exploit middleware approaches
• file systems implemented in middleware.
• Some frustration with the nature of the POSIX
  file interface.

4
Technology

• Talk by STK about tape. (B. Raymond)
• Disk
• Discriminate by market, not by interface
• personal, Enterprise.
• Disk qualities differ depending on market
• Talk by D. Anderson (Seagate) about interfaces
  and features in disks.
• Disk becomes tape

5
Its not Just the Interface!
A drive, is a drive, is a drive.
Or is it ?
Cost
Reliability
Scalability
Performance
Data Integrity
Power Consumption
Desktop 7,200 rpm ATA/SATA
Enterprise 15,000 rpm SCSI/FC
System Architecture
Whats important to the Application ?
6
PS Drive RV Operating Range
7
Application -- Backup

• Current
• Large permanent store people and backup community
  share tape as a media
• Talks by COPAN, Kai Li (Princeton)
• Backup community is interested in disk.
• Accelerated backup
• Diminishes the role of removable media

8
Application Permanent Store

• High Energy Physics sites currently have 2PB.
• Ingest rate to permanent stores are 100 TB/month
• Permanent stores are not necessarily archives.
• Loss rate per replication
• About 1/million

9
Large Permanent Stores

• Currently implemented on tape.
• Requirements beyond naïve cost, capacity
• Security requirements, accident
• Full understanding of errors in the systems
• Issue of continuously ingesting materials.
• Archival quality of personal disk
• Not only Disk BER, block-level logic errors
• FNAL commodity disk/ Linux buffering experience
• Archival disk format should be
• Robust Standard

10
Application -- Buffering

• secure community v.s. high perf/science
  community.
• Commodity v.s. enterprise approach
• Middleware v.s. kernal
• Questions of if posix access semantics are
  sufficient.
• Extensions, if any were not clear

11
Tentative, Personal Gap Analysis from DLSS.

• Disks as a technology
• Big Science buffers are not a market center for
  personal disks
• Analogous to good cluster building?
• In the limit COTS !gt JIT?
• Access
• Evident need to understand if POSIX semantics are
  sufficient.
• Extensions, common basis for them were not clear
• Could be explored
• Permanent Store, as we know it.
• There is Potential for fiducially permanent store
• The kind a program manager would trust
• Work needed, not explicitly supported in the
  market today.
• Backup
• Is increasingly specialized.