Transcript GN3+ - Netmode

GN3plus
What the network should do for clouds?
Christos Argyropoulos
National Technical University of Athens (NTUA)
Institute of Communications & Computer Systems
Network Management & Optimal Design Lab (NETMODE)
(GN3plus - GRNET 3rd Party)
GN3plus Symposium, October, 2013, Vienna
connect • communicate • collaborate
Someone asked for..
• Dynamic provisioning
No-human interaction, fast provisioning/deployment
• Elasticity
Resource upscaling/downscaling on demand
• Mobility/Migration
Rearrangement of resources
• Isolation
Sharing of resources in a multi-tenant environment among users (tenants)
that are unaware of each other profiles
(the ultimate: independence without interference)
connect • communicate • collaborate
1
Clouds started to deliver..
What about networks ?
Need for..
Enhancements..
• Intra – DC (Data Center) networking
Higher Port density
Lower latencies on switching
Programmable Fabric/Flow forwarding
Software switching inside hypervisors
• Inter – DC networking
Layer 2 connectivity
Multi-tenancy
Resource/Service Mobility
Use-cases: load-balancing, disaster recovery
connect • communicate • collaborate
2
Intra – DC
Cloud Requirements
Cloud Computing
Networking
VM provisioning
L2 net interface attachment
VM grouping per tenant
L2 private networking among VMs/tenant
Scale up to millions VMs
Scale up to hundreds of thousands
groups
Scale up to hundreds of thousands L2
private nets
VM mobility
VM net profile portability
Net reconstruction across the Net Fabric
(beyond legacy scripts)
Clouds generate Requests
connect • communicate • collaborate
3
Intra – DC
Network Design Challenges
Networking requests
Networking challenges
L2 net interface attachment
Programmable Access Switches
(at least inside hypervisor)
L2 private networking among VMs/tenant
Programmable Network Fabric (DC
network devices)
Scale up to hundreds of thousands L2
private nets
Multi-bit separators/masks
VM net profile portability
Net reconstruction across the fabric
VM profile binding to Access Switches
Programmable Network Fabric
Requests raise Challenges
Port density
Low latency switching
High availability etc.
connect • communicate • collaborate
4
Intra – DC
Network Design Innovation
Networking challenges
Innovation
Programmable Access Switches
(at least inside hypervisor)
Management Protocol (e.g. NETCONF)
Data Modeling Language (e.g. YANG, NDL)
Programmable Network Fabric
Flow-forwarding (e.g. OpenFlow)
Multi-bit separators/masks
Protocol Stacking *again*
(e.g. VXLAN, NVGRE, FabricPath)
VM profile binding to Access Switches
Programmable Network Fabric
Cloud & Network Management Plane
*Orchestration*
Challenges trigger Innovation
connect • communicate • collaborate
5
Intra – DC
Network Design Solutions
SDN
Innovation
Solutions
Management Protocol (e.g. NETCONF)
Data Modeling Language (e.g. YANG)
Software Switches
(e.g. Open vSwitch/Nexus/JunosV)
Flow-forwarding (e.g. OpenFlow)
Flow controllers
(e.g. Open-source Floodlight Controller,
NEC ProgrammableFlow)
Protocol Stacking *again*
(e.g. VXLAN, NVGRE)
Support for new protocols
(e.g. Open vSwitch, Nexus)
Cloud & Network Management Plane
*Orchestaration*
Unified Control Plane
Innovation generates Solutions
connect • communicate • collaborate
6
Inter – DC
Cloud Requirements
Cloud Computing
Networking
VM grouping across DCs (single slice
across sites)
L2 net connectivity across DCs
VM grouping per tenant
Keep the L2 isolation across DCs
VM migration (cold)
VM net profile portability (e.g. firewalling,
virtual interfaces, MAC mobility, tunnels)
IP mobility
Large data set move
VM migration (hot)
VM net profile portability
IP mobility
Large data set move
Net reconstruction across the fabric
Low latency for memory synch
Clouds generate Requests
connect • communicate • collaborate
7
Inter – DC
Network Design Challenges
Networking requests
Networking challenges
L2 net connectivity across DCs
Distributed Network Fabric (state
synchronization)
Keep the L2 isolation across DCs
Distributed Net Control Plane
VM net profile portability
IP mobility
Large data set move (VM storage)
Distributed Cloud Control Plane
Avoid traffic trombone
High Bandwidth for heavy data flows
VM net profile portability (hot)
Distributed Cloud Control Plane
IP mobility (hot)
Avoid traffic trombone
Large data set move (hot)
Very High Bandwidth for heavy data flows
Net reconstruction across the fabric (hot)
Flow redirection
Low latency for memory synch (hot)
Dark Fibers, Short Distances
Requests raise Challenges
connect • communicate • collaborate
8
Inter – DC
Network Design Innovation
Networking challenges
Innovation
Distributed Network Fabric
Seamless overlay, MAC Multi-homing
(e.g. VPLS, OTV)
Distributed Net Control Plane
Proprietary Net Solutions
(e.g. Google Onix)
Distributed Cloud Control Plane (cold)
Avoid traffic trombone High Bandwidth
for heavy data flows
Proprietary Cloud Solutions (e.g. EC2)
LISP (IP mobility)
Multi-Gig Optical Capacities
Distributed Cloud Control Plane (hot)
Avoid traffic trombone
Very High Bandwidth for heavy data flows
Flow redirection
Dark Fibers, Short Distances
Proprietary Cloud Solutions
LISP
Multi-Gig Optical Capacities
Distributed Hierarchies - Flow controllers
Print money!
Requests trigger Innovation
connect • communicate • collaborate
9
Inter – DC
Network Design Innovation
Innovation
Seamless overlay, Multi-homing
(e.g. VPLS, OTV)
Proprietary Net Solutions
Proprietary Cloud Solutions
IP mobility (e.g. LISP)
Multi-Gig Optical Capacities
Solutions
OpenDaylight
Open vSwitch
…
SDN
Proprietary Cloud Solutions
IP mobility (e.g. LISP)
Multi-Gig Optical Capacities
Distributed Hierarchies - Flow controllers
Print money!
Innovation generates Solutions
connect • communicate • collaborate
10
Open Networking Foundation (ONF)
SDN View
Network Programmability
DC-manager (and authorized
users)
Separate Control-Data plane
L2 topologies
Source: ONF
connect • communicate • collaborate
11
OpenFlow Protocol/API
Flow Forwarding
L2 - L4 flow rules
Source: ONF
connect • communicate • collaborate
12
OpenFlow Switch
Management & Control
Management
interface
Source: ONF
Control
interfaces
connect • communicate • collaborate
13
Open vSwitch
Architecture
connect • communicate • collaborate
14
Source: Linux Foundation
Questions?
Thank You !
connect • communicate • collaborate
15