SURFnet storage pilot plans Rogier.Spoor@SURFnet.nl 17 september - - PowerPoint PPT Presentation

surfnet storage pilot plans
SMART_READER_LITE
LIVE PREVIEW

SURFnet storage pilot plans Rogier.Spoor@SURFnet.nl 17 september - - PowerPoint PPT Presentation

SURFnet storage pilot plans Rogier.Spoor@SURFnet.nl 17 september 2009 What do we do ? OSI-layer Founded to start a nationwide network 1 SURFnet. We make innovation work W hats our activity focus We make innovation work - (online)


slide-1
SLIDE 1

17 september 2009

SURFnet storage pilot plans

Rogier.Spoor@SURFnet.nl

slide-2
SLIDE 2
  • SURFnet. We make innovation work

1

What do we do ?

Founded to start a nationwide network

OSI-layer

slide-3
SLIDE 3
  • SURFnet. We make innovation work

2

W hat’s our activity focus

“We make innovation work”

  • (online) collaboration
  • Thrust
  • Expertise centre
  • Network
  • Authentication and authorization
  • Security
slide-4
SLIDE 4
  • SURFnet. We make innovation work

3

Netw ork got com m odity

  • So when you want to innovate, what’s the

direction?

  • Henry Chesbrough says there’re two options:
  • Up in the stack
  • Down in the stack
slide-5
SLIDE 5
  • SURFnet. We make innovation work

4

Dow n in the stack

  • Digg your own fibers (like Aarnet)
  • Build and develop your own network equipment

Not logical choices

  • Insource the network maintenance

is partly feasible

slide-6
SLIDE 6
  • SURFnet. We make innovation work

5

Up in the stack

Options:

  • TCP or Session level (stack development)
  • Constituent network device management (firewall,

router, LAN)

  • Application layer

Network device management “not done” Only option “the application layer”

slide-7
SLIDE 7
  • SURFnet. We make innovation work

6

W hat’s needed for the application layer?

Foundations are:

  • Computing power (cpu)
  • Storage
  • Network faciliteiten

Architecture:

  • scalability
  • performance
  • reliability (inc. security)
slide-8
SLIDE 8
  • SURFnet. We make innovation work

7

Current application foundation

History based

  • Every service own architecture
  • Every service own physical hardware

Status

  • CPU, Storage en Netwerk not in single architecture
  • CPU Virtualisation SURfnet-services since 2008
  • New SAN storage in 2009 -> for VM
  • Network capacity on demand
slide-9
SLIDE 9
  • SURFnet. We make innovation work

8

Rem ote storage project

  • Small project (in 2007)
  • Two phases
  • Remote storage strategies paper

Phase II

  • Practical test
  • How to outsource storage?
  • Use of lightpaths versus IP
  • Long distance (500 km)
  • Only ethernet (no FC in backbone)
  • Ethernet -> FC converters
slide-10
SLIDE 10
  • SURFnet. We make innovation work

9

How to satisfy storage needs

slide-11
SLIDE 11
  • SURFnet. We make innovation work

10

VM environm ent

slide-12
SLIDE 12
  • SURFnet. We make innovation work

11

Storage costs

http://blog.backblaze.com/2009/09/01/petabytes-on-a-budget-how-to-build-cheap-cloud-storage/

slide-13
SLIDE 13
  • SURFnet. We make innovation work

12

Storage type depending needs

  • NAS/ SAN storage suitable for VM’s and DB’s
  • Cloud storage suitable for archiving data of

streaming media

  • Clustered storage research
slide-14
SLIDE 14
  • SURFnet. We make innovation work

13

New storage project

  • Combine different storage techniques
  • Store VM images, media content
  • Offer a single mount point
  • Easy scalable (multiple peta)
  • Low cost/ petabyte

Start project: Q4

slide-15
SLIDE 15
  • SURFnet. We make innovation work

14

Proposed design

  • DRBD as “SAN/ NAS”
  • Mogile/ Castor cloud storage
  • Fuse switching layer
slide-16
SLIDE 16

DRBD

  • Distributed Replicated Block Device
  • Raid 1 mirror across network
  • Replication over tcp/ ip
  • SuperSockets for replication with dolphin
  • Sync or asynch replication
  • Split brain detection and recovery
  • Open source
  • SURFnet. We make innovation work

15

slide-17
SLIDE 17

Castor

  • Hardware agnostic
  • Massively scalable
  • High performance
  • Guaranteed data integrity
  • Self configuring
  • Self managing
  • Self healing
  • Easy HTTP-based interface

Disadvantage:

  • Difficult to manage files inside castor
  • Commercial software ($$$)
  • SURFnet. We make innovation work

16

slide-18
SLIDE 18

Mogile

  • pen source distributed filesystem
  • http api
  • automatic file replication
  • no single point of failure
  • Filesystem agnostic
  • cheap storage
  • SURFnet. We make innovation work

17

slide-19
SLIDE 19

Glue

Intelligent switching:

  • IO to specific file
  • File extension (db, vm)
  • single mount point
  • Namespace virtualization

Proposed solution: fuse implementation

  • SURFnet. We make innovation work

18

slide-20
SLIDE 20
  • SURFnet. We make innovation work

19

Questions/Suggestions ?