Converge without Conflation

Download Report

Transcript Converge without Conflation

Convergence
without
Conflation
The three-slide statement
Adrian Farrel
Old Dog Consulting
[email protected]
Old Dog Consulting
Misconceptions
• “CAPEX is not an issue”
– Everything is relative!
– Dark fibre may be cheap, but optical switches are not a commodity
• “There can be plenty of capacity”
– There may be plenty of dark fibre, but it is dark
– We are not time travellers and we need to offer bandwidth now
• “Core networks are always over-provisioned and do not need BoD”
– True, they may not need to offer BoD, but:
• How does the network remain over-provisioned?
• “Network operation cannot be automated”
– Ploughing with horses is also very nice!
– Automation is the only way to drive down OPEX
• “There are end-users and network providers”
– There is actually a food chain
– Everyone (except maybe the end-user) is multi-homed
– Transport network operators support (and compete for) multiple access
network providers
• This means a “user” of a transport network may be very large
(for example, a multi-national enterprise), and a change in
capacity requirement can be a big thing.
• This is a layering problem not a peering problem
2
Old Dog Consulting
Requirements
• Requirements in the IETF
– Driven by service providers
– Dynamically change the connectivity between routers or between edges
• “Connectivity” includes capacity
• Assumptions/requirements
– Changes in client network configuration have dramatic effects on server
network load
– Providers need to be able to respond rapidly (i.e., minutes) to new
customer requests
• Truck-roll to turn up new lambdas is not fast
– Client ‘greed’ will be mitigated by server ‘cost’
• But server still does not trust the client!
• Server must retain full policy and operations control
• Need to support prioritised access to resources (pre-emption)
• Virtualisation is a benefit (tends to be connection-oriented)
– Virtual private connectivity
• Pseudowires
• Layer One VPNs
– Virtual router adjacencies
• Layer 3 tunnels
• Transport connections
3
Old Dog Consulting
Solution Toolkit
•
Functional decomposition
– Control plane (rapid provisioning and repair – GMPLS)
– Path computation (via Path Computation Element – PCE)
– Network layering
• Virtual Network Topology (VNT)
•
Operator oversight
– Policy
– Management control (VNT Manager – VNTM)
– Integration with service provisioning
• Understanding of QoS and other service admission control issues
•
•
The IETF will continue to work on this so long as there is service provider
demand
For more details see the full slide-set
•
Why is this less interesting?
– Because the technical problems have already been solved
•
What research work could be done?
– We need to know how stable this type of network will be
– We need to know what the cost savings are
– We need to know how well connected a mesh network needs to be to
provide a good non-blocking ratio
4
Old Dog Consulting
Convergence
without
Conflation
The full slide-set
Adrian Farrel
Old Dog Consulting
[email protected]
Old Dog Consulting
Bandwidth on Demand
What is the IETF Working On?
• Control planes
– Essential for rapid provisioning and repair
– Not fundamental to BoD
– IETF has IP routing, MPLS, MPLS-TE, and GMPLS
• Technology-specific control plane extensions
• Functional decomposition
– Path computation
– Network planning
– Network operation and policy
• Recognition that “on-demand” is really “onrequest”
– Server network must retain control of its own
resources
6
Old Dog Consulting
Definitions and Scope
• A domain is defined as:
Any collection of network elements within a common sphere of
address management or path computational responsibility [RFC
4726 and RFC 4655]
– Classic examples are IGP Areas and ASes
– Equally applicable to technology or client/server layers
• A layer is defined as
separations of technologies (e.g., packet switch capable (PSC), time
division multiplex (TDM), or lambda switch capable (LSC)) [RFC
3945]
– Sometimes called regions
separation of data plane switching granularity levels (e.g., VC4, or
VC12) [RFC 5212]
– Sometimes called sub-layers
a distinction between client and server networking roles [RFC 5212]
• The Traffic Engineering Database (TED)
– The sum of information about the connectivity in a domain (nodes and
links)
– Link constraints (available bandwidth, cost, etc.)
– Configured or learned from distribution protocols
7
Old Dog Consulting
Convergence and Bandwidth on Demand
• Commercial motivators
• Shared server networks resources
– Reduced CAPEX
– Simultaneous support for more client services
– Rapid response to new customers
– A new, marketable service
• Integrated network operation
– Reduced OPEX
– Less steep learning curve
– Protocol robustness
8
Old Dog Consulting
What is a Virtual Network Topology?
• Links in a network may be physical
• Or they may be ‘tunnels’ across a lower layer network
• Tunnels can be configured as services
– Virtual private wire
• The virtual network topology can be tuned on-demand
– Management or control plane operation
9
Old Dog Consulting
Operator Issues - Conflation
• Some service providers are really not happy!
• “On-demand” sounds like the client is in control
– “Customers are not clever”
– “The server resources belong to me”
• “Dynamic” sounds like it might flap
– Transport resources need seconds to provision
• Traditional set-up times of days
• Typical hold-times of weeks
– Client dynamics can be very fast
• Client and server granularities are different
– Smallest server granularity may be 2.5Gb
– Client may deal in micro-flows
• Does not suit all topologies
– “I have laid the fibre so I might as well provide all the bandwidth”
10
Old Dog Consulting
Solution Toolkit
•
•
•
•
Functional decomposition
Policy
Management control
Integration with service provisioning
11
Old Dog Consulting
Path Computation Element (PCE)
– “An entity (component, application, or network node)
that is capable of computing a network path or route
based on a network graph and applying
computational constraints” (RFC4655)
• PCE is a path computation element that
specializes in complex path computation on
behalf of its path computation client (PCC)
– PCE can be:
• Embedded in an NMS
• A dedicated server
• Functional component of a switch/router
• PCEs collect TE information (the TED)
– They can “see” within the domain
12
Old Dog Consulting
Multi-Layer Path Computation
•
•
End-to-end path is not just providing a path across a server network
A single PCE cannot compute a multi-domain or multi-layer path
–
–
–
–
–
•
•
Why not?
By definition of “domain” (unless the PCE can see all layers)
Layers may be commercially separate
Mixing layer information may “confuse” client layer routing
Combining layers might not scale
How to decide which layer boundary points to use?
PCEs can cooperate to derive the best end-to-end path
– Techniques exist for peer domains and can be applied to layers
• Backward Recursive Path Computation (BRPC)
•
But the server layer wants to retain control of expensive resources
13
Old Dog Consulting
VNT Manager Interactions with PCE
• VNT Manager is a policy/management component
• Acts on triggers (operator request for a client TE link, client network
traffic demand info, client TE link usage info, client path computation
failure notification)
• Uses PCE to determine paths in lower layer
• Uses management systems to provision LSPs and cause them to be
advertised as TE links in the client layer
2. I can’t find a path
1. Compute a path
6. Path computation request
and response
PCE
3. I failed to
compute a
path
VNTM
4. Compute a
path
PCE
5. Provision an
LSP and make a
TE link
14
Old Dog Consulting
Integration with Policy
• Policy is fundamental to PCE
–
–
–
–
–
What should a PCC do when it needs a path?
What should a PCE do when it gets a computation request?
Which algorithms should a PCE use?
How should PCEs cooperate?
What should a PCE do when it can’t find a path?
• Note VNTM requests server layer paths NOT client PCE
– Can we set up virtual links ahead of requirement?
– When can we tear down a virtual link?
– Who is allowed to request what type of link?
• RACF PD-FE is a policy component that could use PCE
• Inter-domain paths are subject to Business Policy
– IPsphere Forum is working on business boundaries
• Business policy may guide PCE in its operation
• Selection of domains based on business parameters
is a path computation that PCE could help with
15
Old Dog Consulting
•
Service Management
ITU-T’s Resource and Admission Control Function (RACF)
– Plans and operates network connectivity in support of services
•
Policy Decision Functional Entity
– Examines how to meet the service requirements using the available resources
•
Transport Resource Controller Functional Entity
– Provisions connectivity in the network (may use control plane)
Service control functions
Service stratum
Transport stratum
Network attachment
control functions
RACF
PD-FE
RACF
PD-FE
PCE
PCE
VNTM
TRC-FE
TED
CPE
CPN
PE-FE
TED
TRC-FE
PE-FE
PE-FE
PE-FE
Core network
Access network
16
Old Dog Consulting
Choosing a Server Layer
• Previous consideration is multiple clients
• Client might be connected by multiple servers
• Need to select a server that provides connectivity to the right client
site
• Problem can be seen as VPN membership
• BUT connectivity isn’t the only issues
– Bandwidth
– Quality of Service
– Price
• Problem is similar to selecting a multi-AS path
• Simply solved by client PCE consulting multiple server VNTMs or
PCEs
Client Site 1
Client Site 2
Client Site 3
Server Network 1
Server Network 2
17
Old Dog Consulting
Cascaded Server Layers
Client Site 1
Client Site 2
Client Site 3
• Much more complex to plan end-to-end routes
• Server network could “hide” complexity
• Could use a coordinating PCE
– Hierarchical PCE
• Try to resist “TE abstraction”
• Must enter and leave technology layers in the
same way!
18
Old Dog Consulting
Network Reoptimisation
• Forgotten element of BoD
• Server network may become a mess!
– Incremental addition and removal of services
– Parallel, partially used virtual links
– Minimally used high-capacity resources
• We want to reoptimise the server layer, but:
– Need to consider the impact on the clients
• What traffic can be re-groomed?
• What traffic can be re-routed?
• What bandwidth will be demanded again soon?
• Server layer re-optimisation needs to be holistic
– Optimising individual paths is rarely efficient
• Network optimisation needs to be holistic
– Optimise client and server layers as one function
• This is the job of an off-line planning tool
19
Old Dog Consulting
IETF References
• The IETF is the main originating body for PCE and
VNTM
– PCE working group home page
http://www.ietf.org/html.charters/pce-charter.html
– RFC 4655 A Path Computation Element (PCE)-Based
Architecture
– RFC 5212 Requirements for GMPLS-Based Multi-Region and
Multi-Layer Networks
– draft-ietf-pce-inter-layer-frwk-09.txt Framework for PCE-Based
Inter-Layer MPLS and GMPLS Traffic Engineering
– draft-ietf-pce-brpc-09.txt A Backward Recursive PCE-based
Computation (BRPC) Procedure To Compute Shortest
Constrained Inter-domain Traffic Engineering Label Switched (to
become RFC 5441)
– draft-ietf-pce-global-concurrent-optimization-08.txt Path
Computation Element Communication Protocol (PCEP)
Requirements and Protocol Extensions In Support of Global
Concurrent Optimization
20
Old Dog Consulting
Other References
• The IPsphere Forum can be found at
http://www.ipsphereforum.org
• The ITU-T has worked on several relevant
documents
– Access documents via
http://www.itu.int/publications/sector.aspx?sector=2
• G.7715.2 ASON routing architecture and requirements for
remote route query
• Y.2111 Resource and admission control functions in Next
Generation Networks
21
Old Dog Consulting
Questions
[email protected]
22
Old Dog Consulting