Modular Layer 2 InOpenStack Neutron

Robert Kukura, Red HatKyle Mestery, Cisco

1. I’ve heard the Open vSwitch and Linuxbridge Neutron Plugins are being deprecated.

2. I’ve heard ML2 does some cool stuff!3. I don’t know what ML2 is but want to learn

about it and what it provides.

What is Modular Layer 2?A new Neutron core plugin in Havana• Modular

o Drivers for layer 2 network types and mechanisms - interface with agents, hardware, controllers, ...

o Service plugins and their drivers for layer 3+

• Works with existing L2 agentso openvswitcho linuxbridgeo hyperv

• Deprecates existing monolithic pluginso openvswitcho linuxbridge

Motivations For aModular Layer 2 Plugin

Before Modular Layer 2 ...

Neutron Server

Open vSwitch Plugin

OR

Neutron Server

Linuxbridge Plugin

OR ...

Before Modular Layer 2 ...

Neutron Server

Vendor X Plugin

I want to write a Neutron Plugin.

But I have to duplicate a lot of DB, segmentation, etc. work.

What a pain. :(

ML2 Use Cases

• Replace existing monolithic pluginso Eliminate redundant codeo Reduce development & maintenance effort

• New featureso Top-of-Rack switch controlo Avoid tunnel flooding via L2 populationo Many more to come...

• Heterogeneous deploymentso Specialized hypervisor nodes with distinct network

mechanismso Integrate *aaS applianceso Roll new technologies into existing deployments

Modular Layer 2 Architecture

The Modular Layer 2 (ML2) Plugin is a framework allowing OpenStack Neutron to simultaneously utilize the variety of layer 2 networking technologies found in complex real-world data centers.

What’s Similar?

ML2 is functionally a superset of the monolithic openvswitch, linuxbridge, and hyperv plugins:

• Based on NeutronDBPluginV2

• Models networks in terms of provider attributes

• RPC interface to L2 agents

• Extension APIs

What’s Different?

ML2 introduces several innovations to achieve its goals:

• Cleanly separates management of network types from the mechanisms for accessing those networkso Makes types and mechanisms pluggable via driverso Allows multiple mechanism drivers to access same

network simultaneouslyo Optional features packaged as mechanism drivers

• Supports multi-segment networks

• Flexible port binding

• L3 router extension integrated as a service plugin

ML2 Architecture Diagram

Neutron Server

ML2 Plugin

Type Manager Mechanism Manager

API Extensions

GR

E

TypeDriver

Arista

VLA

N

TypeDriver

VX

LAN

TypeD

river

Cisco N

exus

Hyper-V

L2 P

opulation

Linuxbridge

Open

vSw

itch

Tail-F NC

S

Multi-Segment Networks

VXLAN 123567

physnet1 VLAN 37 physnet2 VLAN 413

VM 1 VM 2 VM 3

● Created via multi-provider API extension● Segments bridged administratively (for now)● Ports associated with network, not specific segment● Ports bound automatically to segment with connectivity

Type Driver APIclass TypeDriver(object): @abstractmethod def get_type(self): pass

@abstractmethod def initialize(self): pass

@abstractmethod def validate_provider_segment(self, segment): pass

@abstractmethod def reserve_provider_segment(self, session, segment): pass

@abstractmethod def allocate_tenant_segment(self, session): pass

@abstractmethod def release_segment(self, session, segment): pass

Mechanism Driver APIclass MechanismDriver(object): @abstractmethod def initialize(self): pass

def create_network_precommit(self, context): pass

def create_network_postcommit(self, context): pass

def update_network_precommit(self, context): pass

def update_network_postcommit(self, context): pass

def delete_network_precommit(self, context): pass

def delete_network_postcommit(self, context): pass

def create_subnet_precommit(self, context): pass

def create_subnet_postcommit(self, context): pass

def update_subnet_precommit(self, context): pass

def update_subnet_postcommit(self, context): pass

def delete_subnet_precommit(self, context): pass

def delete_subnet_postcommit(self, context): pass

def create_port_precommit(self, context): pass

def create_port_postcommit(self, context): pass

def update_port_precommit(self, context): pass

def update_port_postcommit(self, context): pass

def delete_port_precommit(self, context): pass

def delete_port_postcommit(self, context): pass

def bind_port(self, context): pass

def validate_port_binding(self, context): return False

def unbind_port(self, context): passclass NetworkContext(object): @abstractproperty def current(self): pass

@abstractproperty def original(self): pass

@abstractproperty def network_segments(self): pass

Port Binding• Determines values for port’s binding:vif_type and

binding:capabilities attributes and selects segment

• Occurs when binding:host_id set on port or existing valid binding

• ML2 plugin calls bind_port() on registered MechanismDrivers, in order listed in config, until one succeeds or all have been tried

• Driver determines if it can bind based on:o context.network.network_segmentso context.current[‘binding:host_id’]o context.host_agents()

• For L2 agent drivers, binding requires live L2 agent on port’s host that:o Supports the network_type of a segment

of the port’s networko Has a mapping for that segment’s

physical_network if applicable

• If it can bind the port, driver calls context.set_binding() with binding details

• If no driver succeeds, port’s binding:vif_type set to BINDING_FAILED

class PortContext(object): @abstractproperty def current(self): pass

@abstractproperty def original(self): pass

@abstractproperty def network(self): pass

@abstractproperty def bound_segment(self): pass

@abstractmethod def host_agents(self, agent_type): pass

@abstractmethod def set_binding(self, segment_id,

vif_type, cap_port_filter): pass

Havana Features

Type Drivers in Havana

The following are supported segmentation types in ML2 for the Havana release:

● local● flat● VLAN● GRE● VXLAN

Mechanism Drivers in Havana

The following ML2 MechanismDrivers exist in Havana:

● Arista● Cisco Nexus● Hyper-V Agent● L2 Population● Linuxbridge Agent● Open vSwitch Agent● Tail-f NCS

BeforeML2 L2 Population MechanismDriver

Host 1

Host 2

Host 3Host 4

Host 1

VM A

VM G VM E VM D

VM B

VM C

VM F

VM H

VM I

“VM A” wants to talk to “VM G.” “VM A” sends a broadcast packet, which is replicated to the entire tunnel mesh.

WithML2 L2 Population MechanismDriver

Host 1

Host 2

Host 3Host 4

Host 1

VM A

VM G VM E VM D

VM B

VM C

VM F

VM H

VM I

The ARP request from “VM A” for “VM G” is intercepted and answered using a pre-populated neighbor entry.

Proxy Arp

Traffic from “VM A” to “VM G” is encapsulated and sent to “Host 4” according to the bridge forwarding table entry.

Modular Layer 2 Futures

ML2 Futures: Deprecation Items

• The future of the Open vSwitch and Linuxbridge pluginso These are planned for deprecation in Icehouseo ML2 supports all their functionalityo ML2 works with the existing OVS and Linuxbrige

agentso No new features being added in Icehouse to OVS

and Linuxbridge plugins• Migration Tool being developed

Plugin vs. ML2 MechanismDriver?

• Advantages of writing an ML2 Driver instead of a new monolithic plugino Much less code to write (or clone) and maintaino New neutron features supported as they are addedo Support for heterogeneous deployments

• Vendors integrating new plugins should consider an ML2 Driver insteado Existing plugins may want to migrate to ML2 as well

ML2 With Current Agents

Neutron Server

ML2 Plugin

Host A

Linuxbridge Agent

Host B

Hyper-V Agent

Host C

Open vSwitch Agent

Host D

Open vSwitch Agent

API Network

● Existing ML2 Plugin works with existing agents

● Separate agents for Linuxbridge, Open vSwitch, and Hyper-V

ML2 With Modular L2 Agent

Neutron Server

ML2 Plugin

Host A

Modular Agent

Host B

Modular Agent

Host C

Modular Agent

Host D

Modular Agent

API Network

● Future direction is to combine Open Source Agents

● Have a single agent which can support Linuxbridge and Open vSwitch

● Pluggable drivers for additional vSwitches, Infiniband, SR-IOV, ...

ML2 Demo

What the Demo Will Show

● ML2 running with multiple MechanismDrivers○ openvswitch○ cisco_nexus

● Booting multiple VMs on multiple compute hosts● Hosts are running Fedora

● Configuration of VLANs across both virtual and physical infrastructure

ML2 Demo SetupHost 1 Host 2

Cisco Nexus Switch

eth2/1 eth2/2

eth2

nova computenova api

...

neutron server neutron ovs agent

neutron dhcp neutron l3 agent

nova compute neutron ovs agent

eth2

br-eth2

br-int

br-eth2

br-int

vm1

VLAN is added on the VIF for VM1 and also on the br-eth2 ports by the ML2 OVS

MechanismDriver.

The ML2 Cisco Nexus

MechanismDriver trunks the VLAN

on eth2/1.

vm2

VLAN is added on the VIF for VM2 and also on the br-eth2 ports by the ML2 OVS

MechanismDriver.

The ML2 Cisco Nexus

MechanismDriver trunks the VLAN

on eth2/2.

VM1 can ping VM2 … we’ve successfully

completed the standard network

test.

Questions?