WAN - trends and use cases

30
VMX Update

Transcript of WAN - trends and use cases

VMX  Update

Virtualization  concepts

Hardware  Virtualization

• Guest  Virtual  Machines  run  on  top  of  a  Host  Machine  

• Virtual  machine  acts  like  a  real  computer  with  an  operating  system  and  devices• Virtual  hardware  – CPUs,  Memory,  I/O

• The  software  or  firmware  that  creates  a  virtual  machine  on  the  host  hardware  is  called  a  hypervisor

HYPERVISOR

Virtualization  types

• Guest  OS  is  not  modified.  Same  OS  is  spun  as  a  VM

• Guest  OS  is  not  aware  of  virtualization.  Devices   emulated  entirely.  

• Hypervisor  need  to  trap  and  translate  privileged  instructions

Fully  Virtualized

• Guest  OS  is  aware  that  it  is  running  in  virtualized  environment

• Guest  OS  and  Hypervisor  communicate  through  “hyper  calls”  for  improved  performance  and  efficiency

• Guest  OS  uses  a  front-­end  driver  for  I/O  operations

• Example  :  Juniper  vRR,  vMX

Para  Virtualized

• Virtualization  aware  hardware  (processors,  NICs  etc)

• Intel  VT-­x/VT-­d/vmdq,  AMD-­V

• Example:  Juniper  VMX

Hardware  assisted

VMX  Overview

VMX

vCP

vFP

VMX

vCP

vFP

VMX  overview

CP

FP

MX x86  server

VMX

vCP

vFP

Virtual  and  Physical  MX

PFE vPFEMicrocode

TRIO x86

CONTROL  PLANE

DATA  PLANE

ASIC

PLATFORM

VMX  Product

• Virtual  JUNOS  to  be  hosted  on  a  VM• Follows  standard   JUNOS  release  cycles

• Hosted  on  a  VM,  Bare  Metal,  Linux  Containers• Multi  Core• SR-­IOV,   virtIO,   vmxnet3,   …

VCP(Virtualized Control Plane)

VFP(Virtualized Forward Plane)

vMX Product  Overview

VCPVFP

Physical  NICs Management  traffic

Guest  VM  (Linux) Guest  VM  (FreeBSD)

Hypervisor:  KVM,  ESXi

Cores Memory

Bridge  /  vSwitch

Physical  layerPCI  Pass  through  SR-­IOV

VirtIO

Virtual  Control  Plane  (VCP)• JUNOS  hosted  in  a  VM.  Offers  all  the  capabilities  

available  in  JUNOS• Management  remains  the  same  as  physical  MX• SMP  capable  

Virtual  Forwarding  Plane  (VFP)• Virtualized  Trio  software  forwarding  plane.  Feature  

parity  with  physical  MX.  Utilizes   Intel  DPDK  libraries• Multi-­threaded  SMP  implementation  allows  for  

elasticity• SR-­IOV  capable  for  high  throughput  • Can  be  hosted  in  VM or  bare-­metal

Orchestration• vMX instance  can  be  orchestrated  through  OpenStack

Kilo  HEAT  templates• Package  comes  with  scripts  to  launch  vMX instance

VMX  DETAILS

CENTER  CHIP  (MQ,  XM,..)

VMX  Forwarding  Model

LOOKUP  CHIP  (LU,  XL…) Queuing  Chip  (QX,  XQ

,..)

FORWARDING  WITH  TRIO  ASICS  on  MX

DPDK

RIOT

DPDK

FORWARDING  WITH  x86  on  VMX

VMX  Detailed  View

Physical  nics

Virtual  nics

DPDK

Internal  Bridge

172.16.0.3/16vfp-­‐int eth1   :  

172.16.0.2/16

em1:  172.16.0.1/16vcp-­‐int

rpd chasd

VMXT

RIOT

External  Bridgex.x.x.y/m

eth0   :  x.x.x.b/m

fxp0:  x.x.x.a/m

vfp-­‐ext

vcp-­‐ext

vCP

vFP

dcd

DPDK

Using  VMX:  SRIOV  Mode

Physical  nics

Virtual  nics

VCP

VFP

eth0 eth1 eth2 eth3

0 1 2 3

eth0:  vf 0

ge-­‐0/0/0

eth1:  vf 0 eth2:  vf 0 eth3:  vf 0

ge-­‐0/0/1 ge-­‐0/0/2 ge-­‐0/0/3

VFP  ports

JUNOS   portsvCP

vFP

Using  VMX:  Virt-­IO  Mode

Input  can  be  physical  or  virtual

Virtual  nics

VCP

VFP

0 1 2 3

Virtio-­‐0

ge-­‐0/0/0

Virtio-­‐1 Virtio-­‐2 Virtio-­‐3

ge-­‐0/0/1 ge-­‐0/0/2 ge-­‐0/0/3

VFP  ports

JUNOS  ports

vCP

vFP

Using  VMX:  Virt-­IO  Mode

VCP1

VFP1

0 1 2 3

ge-­‐0/0/0 ge-­‐0/0/1 ge-­‐0/0/2 ge-­‐0/0/3vCP

vFP

vCP

vFP

VCP2

VFP2

0 1 2 3

ge-­‐0/0/0 ge-­‐0/0/1 ge-­‐0/0/2 ge-­‐0/0/3

VMX  QoSLEVEL-­1 LEVEL-­

2LEVEL-­

3

PORT

SIX

QUEUES

Q0

Q1

Q2

Q3

Q4

Q5

VLAN  1  

VLAN  2  

VLAN  n

High

Medium

Low

§ Port:§ Shaping-­rate

§ VLAN:§ Shaping-­rate§ 4k  per  IFD

§ Queues:§ 6  queues§ 3  priorities

§ 1  High  § 1  medium§ 4  low  

§ Priority  groups  scheduling  follows  strict  priority  for  a  given  VLAN

§ Queues  of  the  same  priority  for  a  given  VLAN    use  WRR

§ High  and  medium  queues  are  capped  at  transmit-­rate

VMX  PERFORMANCE

Revisit:  X86  Server  Architecture

CPU  Socket  0 CPU  Socket  1

Memory Memory

Memory  Controller Memory  Controller

PCI  Controller PCI  Controller

NICs NICs

Core  

Core  

Core  

Core  

Core  

Core

Core  

Core

Core  

Core

Core  

Core

Core  

Core  

Core  

Core  

Core  

Core

Core  

Core

Core  

Core

Core  

Core

vMX Environment

Description Value

Sample  system  configuration Intel  Xeon  E5-­‐2667  v2  @  3.30GHz  25  MB  Cache. NIC:  Intel  82599  (for  SR-­‐IOV  only)

Memory Minimum:  8  GB   (2GB  for  vRE,  4GB  for  vPFE,  2GB   for  Host  OS)

Storage Local  or  NAS

Sample  system  configuration

Sample  configuration  for  number  of  CPUs

Use-­‐cases Requirement

VMX  for up  to  100Mbps  performance Min  #  of  vCPUs:  4  [1  vCPU for  VCP  and  3  vCPUs for  VFP].Min  #  of  Cores:  2  [  1  core  for  VFP  and  1  core  for  VCP].  Min  memory  8G.  VirtIO NIC  only.

VMX  for  up  3G  of  performance Min  #  of  vCPUs:  4  [1  vCPU for  VCP  and  3  vCPUs for  VFP].Min  #  of  Cores:  4  [  3  cores  for  VFP,    1  core  for  VCP].  Min  memory  8G.  VirtIO or  SR-­‐IOV  NIC.    

VMX  for  3G  and  beyond  (assuming  min  2  ports   of  10G) Min  #  of  vCPUs:  5  [1  vCPU for  VCP  and  4  vCPUs for  VFP].Min  #  of  Cores:  5  [  4  cores  for  VFP,    1  core  for  VCP].  Min  memory  8G.   SR-­‐IOV  only  NIC.  

vMX EnvironmentUse-­case   1:  vMX instance   up  to  100Mbps

Min  #  of  vCPUs:  4  [1  vCPU for  VCP  &  3  vCPUs for  VFP]

Min  #  of  Cores:  2  [1  core  for  VCP.  1  core  for  VFP]

Min  memory  8G.  

NIC:  VirtIO is  sufficient

Core  0 Core  1 Core  2 Core  3 Core  4 Core  5 Core  6 Core  7

VCPU  0 VCPU  1

VCP  (Virtual  Control  Plane) VFP  (Virtual  Forwarding  Plane)

JUNOS I/O   – TX  &  RX

VCPU  2

Worker

Host  OS  

CPU  Socket

Use-­case   2:  vMX instance   up  to  3Gbps

Min  #  of  vCPUs:  4  [1  vCPU for  VCP  &  3  vCPUs for  VFP]

Min  #  of  Cores:  4  [  1  core  for  VCP.  For  VFP  assume  2  port  1G/10G  with  a  dedicated  I/O  core,  1  core  for  each  Worker,  1  core  for  Host  Interface  ]

Min  memory  8G.  

NIC:  VirtIO is  sufficient.  SR-­IOV  can  also  be  used.  

Core  0 Core  1 Core  2 Core  3 Core  4 Core  5 Core  6 Core  7

VCPU  0 VCPU  1

VCP  (Virtual  Control  Plane) VFP  (Virtual  Forwarding  Plane)

JUNOSI/O   port  1  TX  &  RX

VCPU  3

Worker

Host  OS  

CPU  Socket

I/O   port  2  TX  &  RX

VCPU  2VCPU  1

Host  Interface

VCPU  0

Host  Interface

vMX EnvironmentUse-­case   3:  >3Gbps   of  throughput  per   instance

Assume  2  port  10G  for  I/O  

Min  #  of  vCPUs:  5  [1  vCPU for  VCP  &  4  vCPUs for  VFP]

Min  #  of  Cores:  5 [  1  core  for  VCP.  For  VFP  assume  2  port  10G  each  with  a  dedicated  I/O  core,  1  core  for  each  Worker,  1  core  for  Host  Interface]

Min  memory  8G.  

NIC:  SR-­IOV  must  be  used

Core  0 Core  1 Core  2 Core  3 Core  4 Core  5 Core  6 Core  7

VCPU  0 VCPU  2

VCP  (Virtual  Control  Plane) VFP  (Virtual  Forwarding  Plane)

JUNOS I/O   port  1  TX  &  RX

VCPU  3

Worker  1

Host  OS  

CPU  Socket

I/O   port  2  TX  &  RX

VCPU  2VCPU  0

Host  Interface

VCPU  3

Worker  2

VCPU  n

Worker  n

VMX  Performance  in  14.1

vFP

vCP

CPU  Socket  0 CPU  Socket  1

Memory Memory

Memory   Controller Memory   Controller

PCI  Controller PCI  Controller

NICs N

ICs

Core   Core  Core   Core  

Core   CoreCore   Core

Core   CoreCore   Core

Core   Core  Core   Core  

Core   CoreCore   Core

Core   CoreCore   Core

4 5 6 7 8 9 10 11

246

810

1214161820

12 13 14 15 16 17

vMX Gbps

Cores

Performance  for  256B  packets

17  Cores16  gbps

VMX  Performance  in  15.1

vFP

vCP

CPU  Socket  0 CPU  Socket  1

Memory Memory

Memory   Controller Memory   Controller

PCI  Controller PCI  Controller

NICs N

ICs

Core   Core  Core   Core  

Core   CoreCore   Core

Core   CoreCore   Core

Core   Core  Core   Core  

Core   CoreCore   Core

Core   CoreCore   Core

4 5 6 7 8 9 10 11

246

810

1214161820

12 13 14 15 16 17

vMX with vHyper

vMX Gbps

CoresPerformance  for  256B  packets

6  Cores20  gbps

vMX Use  Cases

vLNS for  business  or  wholesale  -­ retail

• Separate  vLNS instance  available  for  each• Business  VPN• Retail  ISP

• vLNS sized  precisely  to  serve    required    PPP  and  L2TP  sessions

CPEAggregation

Access Node

PPP

PPPoE

L2TP tunnel

LAC/  vLAC

Wholesale ISP AAA server

Retail ISPAAA server

Internet

vLNS In Data

Centre

vLNS

Peer  Port

PPECore  sideport

CustomerVPN

Retail ISP

SERVICE  PROVIDER  VMX  USE  CASE  –VIRTUAL  PE  (VPE)

DC/CO  Gateway  

vPE

Provider  MPLS  cloudCPE

L2  PE

L3  PE

CPE

Peering

InternetSMBCPE

PseudowireL3VPNIPSEC/Overlay  technology

BranchOffice

BranchOffice

DC  Fabric

vBNG for  BNG  near  CO

vBNGDeployment  Model

SP  Core

vBNG

Internet

OLT/DSLAM

DSL  or  Fiber  CPE

OLT/DSLAM

DSL  or  Fiber  CPE

OLT/DSLAM

DSL  or  Fiber  CPE

Central  OfficeWith  Cloud  Infrastructure

L2  Switch L2  Switch

• Business  case  is  strongest  when  vBNGaggregates  12K  or  fewer  subscribers

Ethernet

Ethernet

Parts  of  a  cloud

§ CGWRCloud  gateway   router  

Could   be  router,   server,  switch

§ SwitchesSwitch  features  and  overlay  technology   as  needed

§ ServersIncludes  cabling  between  servers  and  ToRs,  mapping   of  virtual  instances  to  ports,  core  capacity  and  virtual  machines

3Leaf

SpineSpine

Leaf

Cloud Gateways

vLNSIP address 1.1.1.1

Other VNFsIP address 2.2.2.2

Server-1

KVMge1 ge2 ge3 ge4

Leaf/TOR

NIC1 NIC2

vLNSIP 3.3.3.3

Server-2

KVMge1 ge2 ge3 ge4

Leaf/TORNIC1 NIC2

Other VNFsIP 4.4.4.4

VMX  with  service  chaining  – potential  vCPE use  case  

vMX as  vCPE vFirewall vNATBranch

Office

Switch  

Provider   MPLS  cloud

DC/CO  GW

BranchOffice

Switch

Provider   MPLS  cloud

DC/CO  Fabric  +    Contrail  overlay

vPE

BranchOffice

Switch

CPE  like  functionality   in  the  cloud

L2  PE

L2  PE

PE

Internet

Thank  you