Sunteți pe pagina 1din 28

Hochverfügbarkeit

in Campusnetzen
Für die deutsche Airheads Community

04. Juli 2017, Tino H. Seifert, System Engineer Aruba


Differences between Campus Edge and Campus Core

Campus Edge Campus Core


 In many cases no High-Availability  In most cases High-Availability
needed needed
 If High-Availability needed, always  In most cases no direct access to end
direct connection to end device device
 In almost all cases LACP only  Lots of different High-Availability
available redundancy option options
 Even in High-Availability deployments  Number of networks with „no
a downtime from time to time is downtime“ requirement ist growing
possible  Campus Edge Protocols are also
valid on Core

3
High Availability at the Edge
A short Overview

4
STP with VRRP

VRRP
Master
VRRP – Configuration intensive: switch by switch
Aggregation/Core Backup
configuration and management
– STP prevents loops, but blocks paths -
x x reducing effective bandwidth
x x
Access – Slow reconvergence impact business,
applications and users
x x

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

5
VRRP without STP

VRRP
Master
VRRP
Backup
– Medium Configuration intensive: switch by
Aggregation/Core switch configuration and management, but
no STP configuration
– No STP needed, as each set of VLANs only
Access on one Access Switch

VLAN1 VLAN2 VLAN3


– Fast convergence decided by VRRP

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

6
Access switch stacking

Backplane stacking Front plane stacking


– Aruba 2930M Switch Series – Aruba 5400R Switch Series
– Topologies – Up to 2 chassis
– Ring: up to 10 units – V3 modules only
– Up to 8 physical links per VSF link
– Aruba 3810 Switch Series – VSF-ports
– Topologies –10GbE or 40GbE port aggregations
–Full mesh: up to 5 units
–Ring: up to 10 units – Aruba 2930F Switch Series
– Front-plane stacking up to 4 units
– VSF-ports:
–SFP uplink models: 1GbE
–SFP+ unlink models: 10GbE

7
IRF/VSF simpler, resilient & high performance networks

– Vastly simplifying L2/3 design, configuration


???
and management
– Link aggregation adds resilience and network
bandwidth
IRF IRF – Fast reconvergence enhances business
continuity and user experience

40G server bandwidth

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

8
DRNI - Distributed Resilient Network Interconnect

– Based on DRNI defined in IEEE 802.1AX


specifications. ???

– DR peer devices exchange information


through DRCP*.
– Forwards traffic locally. IPL IPL

– Supports only two DR peer devices.


DRPort DRPort
– Supports only one IPL** which must be an
aggregation interface.

40G server bandwidth

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

* Distributed Relay Control Protocol **Intra-Portal Link


10
High Availability at the Core
A short Overview

11
IRF/VSF simpler, resilient & high performance networks

– Very easy to configure


IRF
– Link aggregation adds resilience and network
bandwidth
– Fast reconvergence enhances business
IRF IRF continuity and user experience
– BUT: what happens in the case of an
software update – ISSU may help, but
realistically ZERO downtime is not possible

40G server bandwidth

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

12
HP IRF
Traditional Layer 2 Fabric
Pros Cons
 Mature technology  Proprietary
 Simple topology  ISSU not always available
(Single IP for management, peering etc)
 Hash algorithms not perfect
 Simplified configuration file
 Scalability
 Single IRF core = single point of failure?

13
CLOS Fabric
• CLOS (physical) network architecture provide edge/core multi-tier design
• Each leaf switch is connected to all spine switches
• Customers may choose to deploy a 2 spine fabric (2 x 40G uplinks) and expand to 4+
spines (4 x 40G uplinks or more) when they require additional bandwidth
• Protocol independent (STP/TRILL/SPB/L3) over the physical fabric

Spine Switches Spine Switches

Leaf Switches Leaf Switches

2 Spine CLOS Fabric 4 Spine CLOS Fabric

14
L3 Fabric with distributed control plane (OSPF/IS-IS/BGP)

• An L3 Core removes STP while still providing end to end connectivity


• Very high resiliency, but only possible if VLANs don’t need to be distributed to
multiple Leaf nodes
• The practice shows, that almost always someone wants to have at least one VLAN on
multiple Leafs
Spine Switches Spine = independent function

L3 Nework
10G/40G
OSPF/IS-IS/BGP
interconnects

Leaf Switches Leaf = IRF Pair

15
L2 Fabric with distributed control plane (TRILL/SPB)

• TRILL/SPB removes STP while still providing a loop free L2 network for east/west traffic
• Distributed control plane (No single point of failure), but lack of control plane
interoperability
• Architecture Neutral (leaf to leaf or spine leaf)
Spine Switches Spine = independent function

L2 TRILL / SPB Fabric


10G/40G Over IS-IS L3 network
interconnects

Leaf Switches Leaf = IRF Pair

16
VxLAN for Campus networks
A short excurse

17
VXLAN and Overlay Networking Introduction
• Virtual Extensible Local Area Network (VXLAN) is a network encapsulation mechanism first
introduced in 2011
• Supports up to 16 million virtual overlay tunnels over a physical layer 2/3 underlay network for L2
network connectivity and multi-tenancy
• VXLAN allows traffic to be load shared across multiple equal cost paths
• Supports both intra-Campus and inter-Campus deployment scenarios
• VXLAN capable device = VXLAN Tunnel End Point (VTEP)

Data Center (DC) 1 Data Center (DC) 2


Inter-DC
Virtual Overlay VXLAN tunnels Extended Over WAN
VM VM VM VM VM VM VM VM VM VM VM VM
VM VM VM VM VM VM VM VM VM VM VM VM
L2 or L3 L3 WAN L2 or L3 Hypervisor
Hypervisor Hypervisor Hypervisor

Physical Underlay Network Physical Underlay Network


Intra-DC

18
VXLAN Deployment With Centralized Control Plane
• VXLAN with centralized control plane (e.g. DCN VSC, IMC with HPE switches)
• Typically a VM or application installed on a server and includes clustering capabilities for High
Availability (HA)
• OVSDB / NETCONF are examples of protocols used between controller and VTEPs to
setup/teardown VXLAN tunnels and share MAC addresses
Network Virtualization
Controller

OVSDB / NETCONF

VM traffic bridged to physical network via overlay VXLAN tunnel

VM VM VM
VM VM VM
Hypervisor 172.16.10.0/24
Hardware VTEP B Physical
Software VTEP A in rack 1 Underlay Network in rack 100 Routers
Layer 3
VM1
10.0.0.2/24
Physical Bare Metal Server
Firewalls 10.0.0.6/24 19
VXLAN Deployment Without Centralized Control Plane
• VXLAN without centralized control plane (e.g. HPE Comware switches)
• VXLAN tunnels can be setup manually (CLI) or dynamically (MP-BGP EVPN)
• CLI Implementation is Vendor proprietary, don‘t expect interoperability
• EVPN is standardized

Traffic between segments via overlay VXLAN tunnel

172.16.10.0/24
Hardware VTEP A Hardware VTEP B
in rack 1 Underlay Network in rack 100
VM VM VM Layer 3 VM VM VM
VM VM VM VM VM VM
Hypervisor Hypervisor
VM1 Bare Metal Physical Physical Bare Metal
10.0.0.2/24 Server 2 Router Firewalls Server3
10.0.0.6/24

20
EVPN MP-BGP as VxLAN control plane
VxLAN traditional MAC address auto-learning challenge

 Tunnel establishment and VXLAN VNI-tunnel mapping require manual configure


 Tradition VXLAN only define the VXLAN encapsulation, MAC + ARP address learning rely on data plane
flooding, which make BUM packets flood throughout all Fabric

Spine Spine

BUM flooding VXLAN BUM flooding


Network
BUM flooding
VXLAN tunnel
establish

Leaf Leaf Leaf

Control plane:EVPN (RFC 7432)


 MP-BGP to establish VXLAN tunnel between VTEP and map VXLAN VNI to Advantage:
• Simple
tunnel • Automated Overlay
 MP-BGP to synchronize MAC+IP between VTEP, so as to reduce broadcast • ECMP
• Standard Based

21
VXLAN architecture

IP LAN/MAN/WAN
VXLAN VNI 45501

VXLAN VNI 45502

VTEP VTEP VTEP VTEP VTEP

Campus 1 Campus 2 Data


Center

VLAN 501 VLAN 501 VLAN 501 VLAN 501 VLAN 501
VLAN 502 VLAN 502 VLAN 502 VLAN 502 VLAN 502

Devices mapped to VLAN using MAC-based VLAN or MAC-authentication 22


HA for Campus Interconnects
A short introduction

23
ADVPN automates secure connectivity

Campus Simple
– Automated zero-touch deployment with IMC
– Reduces configuration steps
Secure
– Flexible support for any IP WAN technology
– Standards-based IPSec encryption
WAN Scalable
– Site-to-site performance for rich media
– Scales to over 30,000 sites

Secure Data Tunnel

Branch Branch 93 percent reduction in configuration steps


2424
Secure Overlay with IPSec VPN
MM/Standby

Headquarter

Secure Overlay setup over any access - MPLS or VPNC


Internet (Cable, DSL, 4G)

Certified Strong Encryption like FIPS

PSK or PKI based VPN setup MPLS INTERNET

Routing overlay with OSPF and Static Routing

Zero Hub configuration when newer Spokes are


added* Branch Branch Branch

MC MC MC

* Roadmap IPSec Tunnel 25


Context-Aware Path Selection
Policy based Path selection based of
MM/Standby • Applications
• Roles, VLANs
Headquarter

Global Load Balancing based of:


VPNC
• Round robin (default)
• Hash
• Session count
• Uplink utilization
MPLS INTERNET
Uplinks can be configured Active-Active or Active-
Backup
Added WAN Compression capability
Branch
Office 365
Business Critical SAP
MC
IPSec Tunnel Lync/SfB
Latency Sensitive Video

Recreational Youtube 26
Recommendation
Everything should be made as simple as possible,
but not simpler.

27
Downtime on software Upgrades

Whenever there is a downtime possible for


IRF software upgrades -> use VSF / IRF!
– Very stable
– Configuration errors less likely
IRF IRF (always remember – most IT outages are
caused by human error)
– Support for high bandwidth requirements
– Doesn‘t conflict with other network
protocols/functions

40G server bandwidth

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

28
No downtime at the Campus CORE

If there is REALLY NO DOWNTIME possible


L3 Core
for the Campus Core -> USE L3 Campus Core
– Wherever possible try to avoid L2 overlay
– If L2 overlay is needed consider number of
IRF IRF VLAN‘s -> small number configure overlay
manually
– In large scale deployments use EVPN

40G server bandwidth

Configuration effort Operational complexity Active link


Legend
Convergence time User/app experience Blocked/standby link

29

S-ar putea să vă placă și