Reliability in SuperJANET 5

1
Reliability in SuperJANET 5

• Roland Trice
• SuperJANET 5 Project
• r.trice_at_ulcc.ac.uk

2
Introduction

• Background to SJ5 project
• Reliability issues
• How might reliability be improved?
• Funding
• Reliability consultation
• Homework

3
Background to SJ5 project

• Requirements analysis Jeremy Sharp
• Views from the community on
• Reliability
• Services
• Bandwidth
• Applications
• Etc etc etc
• Investigations
• Carrier class routers
• Reliability Roland Trice
• Transmission technology Henry Hughes
• Architecture Duncan Rogerson

4
Background to SJ5 project

• Reporting in December
• Network Strategy Workshop
• JCN, JISC
• Basis for funding provision
• No CSR this time around
• Cant assume how much or over what period

5
Background to SJ5 project

• If funding is released
• Procure new bandwidth in 2004
• Procure routers in 2005
• Deploy in 2005
• Hopefully before clearing
• Must by end of 2005 when SJ4 contract ends

6
Background to SJ5 project

• Key features
• Reliability and availability
• Multiple service streams
• End-end delivery
• http//www.ja.net/SJ5

7
Work in progress

• Meeting with UK MANs
• Investigate ways in which UKERNA and RNOs can
  collaborate to increase reliability
• Reliability questionnaire
• Published on SJ5 web page
• Responses coming in
• The more the better
• Informal investigations
• Telco market
• Carrier class routers

8
Reliability

• Seen as key feature of new network
• JANET necessary for core business activities
• End-end reliability needed for emerging services

9
What do we mean by reliability?

• Maximum availability
• Works first time, every time
• Minimum disruption
• Everyone hates down time
• Predictable performance
• Any time of the day
• Any application

10
What do we mean by reliability?

• Everything works for everyone
• From E-mail FTP to voice and video
• Big institutions down to small colleges
• Web browsing to Peta-byte file transfers
• Difficult to provide with on-size-fits-all
  network

11
What has SJ4 achieved?

• A huge improvement on SJ III
• No user affecting faults on core
• Few RN access link failures
• Improvement plans to rectify problems
• No congestion anywhere on the backbone
• Into regional networks
• To UK ISPs
• To Europe
• To the Internet

12
What has SJ4 achieved?

• Freedom to develop new services
• Not continually fire fighting
• New services bring new problems
• Market test
• Needed to prove someone would sell us
• 16 times more bandwidth
• Flexible contract
• Upgrade path
• Improvement plans
• Development infrastructure

13
Youre never alone with Schizophrenia

• Network must be reliable
• SLA failure will displease funding bodies
• Failure of credibility with user community
• We need to develop new services
• Need to add value where ISP cant
• Failure in credibility if we dont

14
Maximise Reliability

• Choose stable OS and keep to it
• Deploy ultra reliable hardware
• Fault tolerant leased lines
• Eliminate single points of failure
• Strict change control
• Avoid complexity
• Keep your sticky mitts off it!
• At the Core, RN and institution

15
Issues-1

• Choose stable OS and keep to it
• Need to deploy new features
• Mix of features causes instability
• Deploy ultra reliable equipment
• Cost
• Current generation of routers inadequate
• Not carrier class

16
Issues-2

• Fault tolerant leased lines
• Cost
• Diverse routing Vs Diversity of supplier
• Sub-contracting
• Duct swapping
• Eliminate single points of failure
• Costly
• Adds complexity

17
Issues-3

• Avoid complexity
• Resilience and/or more adventurous services all
  increase complexity
• Keep your sticky mitts off it!
• Need to develop new services
• At risk periods are unavoidable
• Strict change control
• Does slow down rate of change
• Temptation to cut corners

18
Issues-4

• Change is difficult to implement on an
  operational network
• Multicast, QoS etc taking months or years to roll
  out
• Problems during deployment will affect
  reliability

19
Squaring the circle

• High reliability and leading edge facilities are
  mutually exclusive on a on-size-fits-all network
  infrastructure
• Possibility that JANET fails to deliver the
  required reliability and the desired innovation
  unless the infrastructure is radically different

20
Multiple independent services

• Commodity best-efforts IP service
• Stable
• Always there
• Minimal at risk time
• Stability, stability, stability

21
Multiple independent services

• Offer greater range of leading edge facilities
• Early adoption of emerging technology
• Special application environment
• Guaranteed latency and jitter
• Very high bandwidth
• Managed bandwidth

22
Multiple independent services

• Will almost certainly need more attention
• More frequent at-risk work
• Less stability
• Trickle down
• New services can be brought into production
• Once all the bugs have been ironed out

23
Multiple independent services

• Not just about reliability
• Allows greater exploitation of the network
• Allows support of a diverse community
• Needs of the many are not out-weighed by the
  needs of the few or the one
• Needs of the few or the one are not stifled by
  the needs of the many

24
Provision

• MPLS
• TAG switching of different services
• Still relies on on-size-fits-all network
• Complex
• Untried in our community
• Not flexible enough
• May have a place if no other way can be found
• Potential use in legacy networks

25
Provision

• Multiple virtual links from a Telco
• Lambda or SDH services
• Use optical muxes to apportion bandwidth
• Could provide raw bandwidth without IP
• Very high data rates available
• 10 Gbit/s now
• 40 Gbit/s shortly
• Possibly by the time we procure

26
Provision

• Dark or Managed fibre Fibre
• As above, but we own and operate the fibre and
  optical kit
• Long term cost benefit
• If we can get away from 5 year procurement cycle
• Absolute control
• Maintenance issues

27
Routers, The Next Generation

• NG routers 99.999 availability
• Better hardware
• More reliable
• Very high capacity
• Cunning features
• Caching of routing and forwarding
• ATM like QoS control
• Virtual Routers
• Modular software

28
In the real world

• Core network very reliable, access links are the
  weakest link
• Telco fault handling remains poor
• Lies, damn lies and the art of parking
• Duplication of links
• RN to JANET
• UKERNA RNOs will look at resilience
• Institution to RN
• Institutions need to talk to their RNO

29
In the real world

• Institutions will have to fund duplicate links
• More effective than shouting at a Telco
• Institutional power problems still a significant
  issue

30
Funding, funding, funding

• Everyone says they want more reliability
• Not a free good
• Who wants to pay for more reliability?
• Ah well, err maybe in a few months.
• Need to put a price on reliability
• Cost to institution of a loss of JANET service
• Indicate spending that mitigates risk
• LAN or WAN or both or none? 
• Results could influence the spend if not the
  purse

31
Reliability consultation

• Available now on SJ5 web page
• To determine how important JANET actually is to
  institutions and to users
• To recommend ways in which reliability may be
  improved in a scaleable and affordable way
• Cost benefit analysis to maximize value for money
• Responses by mid November please

32
Key questions

• Dependence on the network
• How tolerant is your institution to outages?
• Have you identified a cost of network down time?
• Most responses dont give a figure
• Investment priority
• LAN - WAN
• Risk Analysis
• Resilience
• Spending

33
Key questions

• Balance between reliability and development
• Regional Network Reliability
• At-Risk sessions
• Institutions Risk Analysis

34
Homework

• Talk to your institution, UKERNA and your RNO
• Respond to the questionnaire
• If possible, with an institutional remit
• Highly desirable
• Risk analysis
• Cost implications
• Better reliability comes at a price
• Cost benefit analysis
• Smart spending

35
Questions