Transcript GN3 Tx and Sw Equipment Procurement Update

New GEANT network:
BoD service migration & NSI v2.0
Guy Roberts 17 Sept 2012
connect • communicate • collaborate
Topics
Current BoD service
New GÉANT network
BoD in new network
NSI v2.0
connect • communicate • collaborate
2
Review of BoD in existing GÉANT network
connect • communicate • collaborate
3
Old network: MCC delivers
GÉANTplus and BoD services
Lambdas
Primary IP
Peering
IP
GÉANT+
Backup
GÉANT+
IP
Backup
To NREN B
To NREN A
IP Router
Primary IP
Peering
Switch
Switch
IP Router
Dark fibre/amplifier chain
GÉANT
2
POP A
GÉANT
2
POP B
connect • communicate • collaborate
Lambdas
Old network: GÉANT Dynamic Core
Alcatel 1678MCC equipment using SDH timeslots (VCGs)
Bandwidth is guaranteed as no capacity sharing
Supports EPL (port-to-port circuits) and EVPL (VLAN-toVLAN circuits)
Due to be replaced as:
Poor user experience due to lack of ingress buffering in
EVPL mode
No statistical multiplexing –
so a high cost solution
connect • communicate • collaborate
5
Current NEXPReS infrastructure on GÉANT Dynamic Core
Dynamic Core
connect • communicate • collaborate
MCC Buffer problem…
•
•
•
•
•
•
•
Use of Iperf on EXPReS
servers
VC-4-48v path
TCP throughput
Zero path length
Max. 100 flows – Iperf limit
Throughput falls with large
TCP window
Bottleneck is managed with
small buffers approx. 64kByte
Attempts to fix this by Alcatel
in later versions were only
partially successful.
8
7
Throughput (Gb/s)
•
TCP window size vs. throughput
6
100 flows
5
line rate
4
10 flows
3
1 flow
2
1
0
1
10
100
1000
10000
TCP window (kbytes)
• Tuning TCP window very hard for single flow!
• Tuning TCP window very hard for single flow!
connect • communicate • collaborate
Sharing dynamic core: vlans
TP is able to accept connection requests from more
than one Domain manager
To share infrastructure for BoD and NSI demo we need
to ensure that conflicts are not possible.
NSI uses vlan range 1780 to 1800, BoD uses 20002050.
Alcatel TP should be modified to validate correct usage
before forwarding to OMS.
connect • communicate • collaborate
8
Future BoD infrastructure
connect • communicate • collaborate
GÉANT Procurement
3 layers in the network being replaced:
Dark fibre
DWDM transmission equipment (Layer 1)
Switching (Layer 2)
Router equipment will initially remain unchanged
Juniper T-series remain
Later routing function will be migrated to new Juniper
MX equipment
connect • communicate • collaborate
10
DF procurement
Current lease on GÉANT DF expires in 2015
To support the migration we decided to re-procure dark fibre
early – having a second pair of fibres during migration makes
process easier and reduces customer outages.
Contracts have been awarded
5 DF providers: Colt, Telia Sonera, Interoute, Level3, Pantel
New lease runs to 2020
connect • communicate • collaborate
11
New DF footprint: up to 2020
connect • communicate • collaborate
12
New optical layer: Infinera
Next-gen
PIC-based
DWDM
Next Generation Integrated
WDM Transport, OTN, &
Packet Switching platform
Up to 8Tb/s capacity on ILS in initial release
Up to 80 x 100G channels with ULH reach
PIC-based 500Gb/s DWDM line cards
Bandwidth
Virtualization
GMPLS &
MPLS-TP
Multi-Tb/s OTN & Control
Plane
Packet Switch
5Tb/s Bandwidth Virtualization in single bay
Fully non-blocking OTN and P-OTN
switching
Multi-layer GMPLS, MPLS-TP
Software intelligence for OTN & packet
services
connect • communicate • collaborate
O-E-O
O-E-O
O-E-O
O-E-O
Digital ROADMs
O
O O O
O-E-O
O-E-O
O-E-O
O
O
O
O-E-O
O
O
OXC
O O O
ODU0 & packet
bandwidth
management
O
O
O
O-E-O
O-E-O
O-E-O
O
O-E-O
OTN & packet services
Infinera Digital Optical Network
Conventional WDM Networks
•
•
•
Separate WDM/ROADM & OTN switching
layers
Sub-l grooming & muxing only with OXC
Manual grooming complexity or extra cost
for OXC
•
•
•
Integrated WDM and OTN/packet
bandwidth management
Sub-l grooming at every node
End-end service management, PM and
OAM
connect • communicate • collaborate
Switching layer: SDH to be replaced
with MPLS
New switching platform uses Juniper MX equipment
to deliver MPLS service
Circuit services access using VLANs on a shared
IP/GÉANT Plus interface.
Capacity for GÉANT Plus up to 100Gbps.
Based on MPLS technology
GÉANT Plus will not be policed at ingress…
some implications here…
connect • communicate • collaborate
15
Day-1 connectivity used by MXs and
extant routers (high level view)
DK
EE
LV
LT
PL
RU
BE
IE
UK
NL
DE
CZ
SK
LU
FR
CH
AT
HU
RO
Fink
PT
ES
IT
SI
HR
BG
KEY
100G lambda
Nx10GE lambda
Nx10GE lambda (detail TBD)
Nx10G leased lambdas
GR
connect • communicate • collaborate
16
Planned new GÉANT architecture
(principles)
100G trunks to accommodate GÉANT IP and GÉANT Plus from day-1
Where a POP has a router and MCC today, replace MCC with MX960
Where a POP has an MCC and no router then replace with two MX480
Sub-10G GÉANT Plus instances implemented as VLLs on MX platform
Existing 10G GÉANT Lambdas will be migrated to Infinera platform
Encourage new 10GE GÉANT “Lambdas” (now really higher capacity
GEANT Plus instances) to be carried over new Lot 2 platform
connect • communicate • collaborate
17
Generic lot 2 and extant IP
architecture after migration (“day 1”)
POP A
POP B
IP
primary
POP C
IP
peers
Logical link: K Gbps (via Lot 2 box)
Router
Logical link: K Gbps (via Lot 2 box)
Logical link: Mx10G (bypassing Lot 2 box)
Mx10G
Router
Router
Mx10G
Nx10G
Nx10G
Nx10G
DF line
Lot 2
switch
DF line
100GE
GÉANT
Lambda
>10Gbps
Lot 2
switch
GÉANT
Plus
100GE
100GE
Lot 2
switch
GÉANT
Lambda
>10Gbps
IP
backup
“IP Peers” = interconnects, commodity upstreams,
real peers, etc
connect • communicate • collaborate
18
Converged lot 2 and IP architecture
at “day-2” (current vision)
By this stage DF lines expected to be carrying multiple 40/100G
lambdas so GÉANT IP backups (or dual-homed IP primaries) can be
transported directly to other POPs avoiding the single Lot 2 switch
POP A
POP B
POP C
IP
primary
DF line
Lot 2++
switch
100GE
DF line
100GE
GÉANT
Lambda
>10Gbps
Lot 2++
switch
100GE
GÉANT
IP Plus IP
peer
backup
100GE
Lot 2++
switch
GÉANT
Lambda
>10Gbps
But need to think a bit more about how to
get from “day-1” to this “day-2” safely
connect • communicate • collaborate
19
A bit about quality of the new GÉANT
Plus service instances
Two consequences of moving to the new “packet transport”:
Absolute BW guarantee is no longer there
Microscopic jitter performance will be “worse”
BW “guarantee” will now done via a CAC process and traffic
engineering
DANTE is also investigating and will purchase commercial
capacity planning/traffic management tool:
GÉANT IP traffic and GÉANT Plus share trunk.
connect • communicate • collaborate
20
AutoBAHN/BoD Tech Proxy
AutoBAHN/BoD requires a unique ‘technology proxy’ for each
technology/vendor combination
DANTE Systems team have developed an Alcatel technology
proxy that allows BoD to automatically create GÉANT Plus type
circuits
A similar Juniper technology proxy is being developed for BoD
Target date for completion November 2012
Service migration date not set yet – expect early 2013
If successful the migration to the new system should be
transparent to the user.
connect • communicate • collaborate
Architecture for BoD tech proxy
connect • communicate • collaborate
22
What should users expect from
GÉANT platform migration?
Migration from Alcatel to Juniper platform should be
transparent to the user.
New platform will allow users to create circuits with
capacities >10Gbps
Fixes current buffer bug in 1678MCC
Juniper expected to have a more robust north-bound
interface to AutoBAHN/BoD
Statistical multiplexing uses bandwidth more efficiently, but
responsibility on operator/policy to manage bandwidth
utilization
connect • communicate • collaborate
23
NSI v2.0
connect • communicate • collaborate
24
NSI v2.0
WSDL, State-machine and topology are now stable
Development has begun for a v2.0 plugfest in OGF36 in
Chicago
NSI Connection Service v2.0 documentation starts now –
however beware, editing process is usually slow
NSI v2.0 demo/plugest will happen at supercomputing 2012
in November.
connect • communicate • collaborate
25
New in NSI v2.0: Primitives and SM
The Reserve, Provision, Release, Terminate, ForcedEnd
and Query messages are carried over from v1.0.
The Modify message has been added to allow a connection
end time to be modified.
New concept of a Message Delivery Layer is introduced in
v2.0. This layer supports confirmation of the delivery of
message to all immediate children including uPA in the
same NSA.
A completely re-worked state machine is defined - supports
modify command
connect • communicate • collaborate
26
New in NSI v2.0: STPs and
Pathfinding
STPs are constituted of 3 parts: <network><localId><label>
<network> identifies a network resource managed by an NSA
<localId> identifies a local resource in the network (eg a port)
<label> is a tuple identifying constraints on technology specific
attributes. Eg, VVLAN=1037, Can also be expressed as lists:
VLAN=1030-1050
Connections requests can now include Explicit Routing Objects
(EROs) in the form of STPs.
Better NML compatibility: Ids now OGF URN compliant
Mechanism to associate NSAs with a topology
connect • communicate • collaborate
27
New in NSI v2.0: Other
NSA version discovery service. New web service in v2.0 allows
queries to an NSA to find information such as the NSI version
and associated SOAP endpoints.
Security: In CS v2.0 the inter-domain transport layer security
utilize TLS with client/server public key certificates. This
enables:
a. Confidentiality (using TLS)
b. Integrity (implicit with point-to-point transport security)
c. Mitigation of Third-Party Replay Attacks (implicit with point-topoint transport security)
d.
Mutual Authentication (client/server public key certificates)
Error handling – enhanced error reporting including hierarchical
reporting of errors in reservation tree.
connect • communicate • collaborate
28
Thank you!
connect • communicate • collaborate
29