welcome to dell emc networking · 18-09-2017 · • macvlan: each container gets an individual...
TRANSCRIPT
- Gartner
Welcome to Dell
EMC Networking
“The world’s most innovative, disruptive &
visionary networking vendor…”
Internal Use - Confidential
2
Internal Use - Confidential
Recognition and Vision from Gartner
2014 2015 2016
“All organizations globally should consider Dell EMC, provided there is appropriate channel and sales coverage with networking skills in the region” – Extract from Gartner Magic Quadrant for Data Center Networking, July 2017
2017
Trends in
Networking
Bernd Kornmaier
Agenda
• Data Center Transformations
• Ethernet Evolution
• In-Rack Networking Trends
• Data Center Fabrics & Interconnect Trends
• Open Networking Software Trends
Data Centers enablers of Digital Transformation
Mobile Apps
Web 2.0
Cloud Services
Big Data
Web
Wireless
Device Explosion
CONTENT ACCESSDATA CENTER
EXPLOSION
Multi-Cloud
Open Source
Virtualization
Containers/Microservices
SDN
NFV
Datacenter Network Must Enable Digital Transformation, Adapt to 3rd Platform
Source: IDC 2017
Traditional and Cloud-Native IT
ITIL-based
IT processes
Client-server
scale-up apps
Infrastructure
resiliency
Coexisting IT paradigms
DevOps based
IT processes
Distributed
scale-out apps
Application
resiliency
Dell EMC is challenging the networking status quo
Proprietary
ASICs
Proprietary
architectures
Proprietary
software
Shifting the paradigm
Business
agility
Investment
protection
Innovation
velocity
Traditional Networking Open Networking
Ethernet Evolution
100GbE & Beyond: Ethernet Trajectory
Total DC Ethernet ports
shipped * Dell Oro Group
Demand Growth * LightCounting
100G Adoption will ramp
faster than 40G * Dell Oro Group
Cloud Early Adoption* LightCounting
In-Rack Networking Trends
100GbE takes off
…Result is 100GbE has
become the new 40GbE
• High Density Single chip solutions – Lower System Costs
• Economies of scale benefitting Optics –Lower Optics Cost
• Switch ASIC process technology evolution –lower Power
• Switch ASIC lookup engine efficiencies –lower Latency
• Backwards Compatible Form Factors
• Higher switch radix - 100GbE ready for Aggregation / Spine
Transition Use Case - 10GbE Down / 100GbE Up
S4
10
0 L
eaf
100GbE Spine
10G
bE
10G
bE
• With onset of affordable 100G in the
Aggregation layers, existing rack
components need to connect to 100GbE
Fabric
• Optimize infrastructure with 10GbE down to
the rack and 100GbE up to the fabric
• Dell has responded with 2 new families:
– S4100 Series
– S4200 Series
25GbE in the Rack
Take advantage of embedded 25GbE
I/O’s shipping in the latest generation
of servers, storage & networking
• Expand use of virtualization
technologies in both servers and
storage
• Grow use of cloud-based and web-scale
technology
• Increase use of flash technology in
storage systems requiring high-speed
I/O
• Increase I/O performance for multi-
processor servers
25GbE
NIC
25GbE
NIC
14G Server
iSCSI Storage Systems
10/25GbE 10/25GbE
3
25GbE
25GbE
10
0G
bE
100GbE
Move to 25GE on Servers is Happening
10/40G
25/100G
Emerging Use Case - 25GbE Down / 100GbE Up
S5
10
0 L
eaf
100GbE Spine
25G
bE
25G
bE
• As soon as 25GbE hits economies of scale (near parity with 10G) -- 25G will become the de-facto standard.
• 25G down and 100G is the coming default build demand. Better utilization of existing components
• Dell Responds with multiple 25G TOR options
– Z9100-ON – 25G Breakout Support
– S6100-ON – 25G Breakout Support
– S5048-ON – Native 25G Support
– S5148-ON – Native 25G Support
Data Center Fabric Trends
Fabric Trends
ARCHITECTURES
Scale Out
Leaf-and-Spine
MANAGEMENT
DevOps
Fabric-wide Mgmt.
SOFTWARE DEFINED
Operational Agility
Software Defined Fabric
CONVERGENCE
RDMA over Converged Ethernet
NVMe over Fabric
AUTOMATION
Fabric Controllers
Workflow Automation
VIRTUALIZATION
Multitenancy
Overlay Networking
Scale out Multitenant Data Center Fabrics (EVPN)• Scale-out Network Virtualization with BGP EVPN VXLAN
• BGP EVPN - Controller-less switch based Ethernet VPN
technology for scale out data center fabrics
• Leverage traditional layer 3 underlay
• BGP control plane for VXLAN overlays
• Inter-tenant Routing (VRFs) and Multi-homing (VLT)
• Benefits
• Open and standards based fabric technology
• Scalable multitenancy
• VM Mobility in the overlay
• Use Cases
• Intra Data Center Network Overlays
• Data Center Interconnects (DCI)
BGP Control Plane
VXLAN Data Plane
IP Underlay
EVPN
EVPN Technology Overview
VTEP-1 VTEP-3 VTEP-4VTEP-1 VTEP-5
iBGP EVPN iBGP EVPN
VLTVLT
IP
BGP EVPN RR BGP EVPN RR
VRRP VRRP
VRF VRF
VTEP-2 VTEP-5
BGP EVPNAS 65001
Configure BGP EVPN address family and IP underlay; VLT for dual homing
Configure VxLAN VNI-to-VLAN map
Configure VRFs and associate VNIs for tenant segmentation
1 2 3
Data Center Interconnects (DCI)
• Interconnecting Data Centers over a Private
Wide Area Network
• Market disruption via merchant silicon based
platforms and disaggregated model
• Traditional technologies: IP Routing, MPLS
• Emerging technologies: BGP EVPN VXLAN
• Public Clouds and Service Providers: Segment
Routing
• Advances in long reach Optical technologies
(100G and Beyond) enablers of traditional
switching for DCI use cases
Fabric Automation Trends
Cloud Based Fabric Design & Deployment• Simplifies and automates new fabric deployments and scale
out of existing fabrics in the data center.
Intent based configuration
• Design driven by specifying business intent
Dell EMC SmartFabric
Dell EMC Fabric Design Center
Faster time to deployment• Plug-and-play fabric deployment
Predictable always-on performance• Automatic topology validation
Fabric-level management• One cli, API, & console to design, build and monitor fabrics
Self-healing fabric
• Automatically detects fabric misconfigurations or link
level failure conditions
Open Networking Software Trends
Broad Open Networking OpportunitiesGreen & Brownfield – In & Around DC – Software the Differentiator
Traditional
Virtualized
DC
SDDCPrivate
Cloud
Network
Packet
Broker
DCIDMZ
Boundary
Application
AnalyticsWAN DevOps
Open
Source
OS9 &
OS10 Enterl l
OS10 Open l* l l
BSN – BCF l l l
BSN – BMF l l
Pluribus l l* l l l l
Cumulus l l* l* l l
IPI l l
Silver Peak l
NSX l l l
Midokura l l
DC
* = when paired with a NVO
Specialized Tools, Software, Licenses Linux & Open Source
Common Hardware
NV
O
Networking OS Macro Trends
SAI Compliant, CPS Programmable OS10
DevOps solutions ready
Single Management pane for heterogeneous compute and Networking Infrastructure
Analytics and Telemetry
Collect and analyze practically everything about Network Fabric
Large Scale out Network Designs using standardized features
Linux Automation Ready
Disaggregation of Networking Stack
Open networking
Optional SDN/NVO controller
Standard orchestration & automation tools
Any networking OS
Open standard hardware
Merchant silicon
Closed Networking
Proprietary ASICs
Proprietary networking OS
Hundreds of protocols
Proprietary architectures & management tools
OS10
Containers and Microservices
• Single-host networking: Private network
within host (NAT for outside)
• Multi-host networking evolution:
• Bridged: Containers connected to
Linux Bridges (L2 virtual switches)
• VXLAN encapsulation between
hosts for L2 network
• MacVlan: Each container gets an
individual MAC and IP in the same
subnet as host
• IPVlan: Each container shares
same MAC as host and individual
IP in any subnet
Linux Containers Networking with Docker
• Software architecture pattern possible with
today’s technologies (containers, network
virtualization, cloud computing)
• Improves process isolation, security,
restartability, and upgradability
• Increased East-West traffic from servers
Microservices
Analytics with In-band Network Telemetry (INT)
• Improve monitoring of P+V network
traffic and fault isolation
• Real time
• Low CPU overhead (Data plane)
• End point (P or V) monitoring
• Latency
• Link utilization
• True paths
• Packet-level granularity
• Push model
• Under the scope of the P4 consortium.
Model sponsored by Dell EMC, VMware,
Intel, Arista and others
In-band Network Telemetry (INT)
How it works
• INT-aware devices insert INT messages
in packets, real-time and transparently
• INT-aware devices can consult the
information and expose data to clients
• INT-aware devices could also react upon
INT data
• INT messages are removed at destination
What is Intent Based Networking
1. Translation and Validation– The system takes a higher-level business policy
(what) as input from end users and converts it to the necessary network
configuration (how). The system then generates and validates the resulting
design and configuration for correctness.
2. Automated Implementation – The system can configure the appropriate
network changes (how) across existing network infrastructure. This is typically
done via network automation and/or network orchestration.
3. Awareness of Network State – The system ingests real-time network status
for systems under its administrative control, and is protocol- and transport-
agnostic.
4. Assurance and Dynamic Optimization/Remediation– The system
continuously validates (in real time) that the original business intent of the
system is being met, and can take corrective actions (such as blocking traffic,
modifying network capacity or notifying) when desired intent is not met.
Why IBN:
“Networks are holding
businesses back as most
enterprise networks are
complex, rigid, hard to
change and slow to
provision, hard-coded
and downright slow to
change any kind of
security defenses *
Forbes.com
“Don’t tell me What to do, Tell me What you want!”
* IBN as defined by Gartner
Open Networking Solutions with Pluribus:
New Feature: Native Visibility with Open DCI
DC-3
DC-1 DC-2
s6010 s6010
s6010
s6048T
s6048Ts6048T VTEPs
VTEPs
VTEPs
L3Underlay
VXLANHATunnel
VXLA
NHA
Tunn
el
VXLANHA
Tunnel
Pluribus Open DCI bridges multiple sites
The same visibility is now everywhere
Data is displayed in the same dashboard
Flow Connection Analytics
‒ Real-time
‒ No Sampling
‒ Correlation between server-client
‒ Correlation between overlay/underlay
Performance Analysis
‒ Intra-DC & Inter-DC
Security/Forensic Analysis
Integration with vCenter, Nutanix
REST API
31 of Y
Internal Use - Confidential
5-Site DCI on
Dell
Hybrid-Fabric w/
E28
Multi-tenancy
Virtual Link
Extension HA
Per Tenant
Connection Flow
Analytics
VTEP HA
Multi-site DC
fabric
Loopfree
overlay
3-Site DCI on
Dell
31
DCI Use Cases With Dell
VTEP HA
Multi-DC Fabric
Nutanix
analytics
2-Site DCI on
DellOpen
DCI
Cloud Service Provider - Itconic
Why Pluribus / Dell:We gained a future-ready network for our cloud
services by working w Dell
Customer Comments - López García explains: “We wanted
to implement a more open solution, specifically on
networking, giving us greater freedom to make changes as
we needed.”
Cisco solutions were “too predefined” and lacked
adaptability. Opinion among key stakeholders was that
Cisco didn’t focus enough on its relationships with partners,
particularly organisations such as VMware, which would
play a central role in Itconic’s cloud services platform
Pluribus Benefits
Virtualization
Controller less SDN
High Scalability
Brownfield implementation
Visibility on what is going
on in the network
Multi-tenancy
Service insertion
Open Networking Solutions with BigSwitch:
Internal Use - Confidential
IDENTIFYING A BIG SWITCH OPPORTUNITY
• Network Refresh
• New Datacenter PODs
• OpenStack
• Cisco ACI Pilot
35© 2014, BIG SWITCH NETWORKS, INC. PROPRIETARY AND CONFIDENTIAL
• Network Security
• Network Monitoring
• Broader Visibility
• Security Service Chaining
BIG CLOUD FABRIC
BIG MONITORING FABRIC
36
Internal Use - Confidential
Shared “ONE BIG SWITCH” ArchitectureDisaggregation of the “MainFrame”
FABRIC CARD
SUPERVISOR(S)
LINE CARD(S)
LINE CARD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
SUPERVISOR
2
FA
BR
IC
CA
RD
FA
BR
IC
CA
RD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
SUPERVISOR
1
FA
BR
IC
CA
RD
FA
BR
IC
CA
RD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
SUPERVISOR
2
FA
BR
IC
CA
RD
FA
BR
IC
CA
RD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
LINE CARD
SUPERVISOR
1
FA
BR
IC
CA
RD
FA
BR
IC
CA
RD
BIG CLOUD
FABRIC
CONTROLLER
HierarchicalControl Plane
1 3
SPINE SWITCHES
2
10G/40G
Backplane
41 32 4
COMPUTE
WORKLOAD
SERVICES &
CONNECTIVITY RACKSCOMPUTE WORKLOAD
LEAF SWITCHES
Physical&
VirtualWorkloads
1G/ 10G/40G
Workloads
• Disaggregated frame – One “Big Switch”
• Open, Simple, Economical, Vendor Choice, Scale-out
• Traditional frame design
• Single point of management
• Proprietary, Expensive, Lock-in, Fixed Slots
NET
NET
NET
SDN & Clos Fabric Necessary for NetFrame Disaggregation
37
Internal Use - Confidential
! tenant
tenant BLUE
logical-router
route 0.0.0.0/24 tenant system
interface segment web
ip address 10.1.1.254/24
segment web
member-port-group pg-bm0 vlan 20
Why Customers Buy: 1) Simplicity
External Core
Router
WEB WEB APP APP DB DB
Segment-Web Segment-DB
MULTIPLE L2 SEGMENTS
Segment-App
Logical Router
(w/ policy)
LB
FWTENANT BLUE
Application Agility(Logical Networking,
Provisioning Templates)
Hitless FabricUpgrade
~15Minutes
Hitless Upgrade(Controller coordinated)
Zero-Touch Fabric(REST APIs, GUI, CLI)
BoxbyBox
Feature Big Cloud
Fabric
Switch OS Install Automatic
Link Aggregation Automatic
Fabric Formation Automatic
Trouble-shooting Fabric-wide
L4-7 Service
Chaining
Declarative (per
tenant)
Add/Remove/
Update Fabric
Automatic
Fabric Visibility Controller or API
BoxbyBox
BoxbyBox
16 racks, 40 devices
38
Internal Use - Confidential
Why Customers Buy: 2) RESILIENCY @ SCALEChaos Monkey Resilience proves BCF is Best in class HA at Scale
Chaos Monkey Testing: 42k simulated
End-points/VMs of background load
and 640+ forced component failures
during the “under stress” test runs
32 leaf / 6 spine / 16 rack pod
Controller fail-over every 30 seconds
Switch fail-over every 8 seconds
Link fail-over every 4 seconds
Conclusion: 640 component failures in 30 minutes with no
impact on application performance
39
Internal Use - Confidential
Why Customers buy: 3) simple, simple, simple
Verify Logical Path
Segment, Logical Router, L3 Policy, L4-7 device (next hop)
View Simulated Topology
Source ingress Leaf Spine egress Leaf Dest
Fabric Trace Fabric Analytics
Easy fine-grain time-series search
of log events based on:
• Event state (e.g. failures)
• Configuration change (REST,
CLI or GUI)
• Tenant / Segment / Devices
• End-point (MAC or IP)
attachment & detachment
Fabric Programmability
– Native REST APIs: GUI & CLI are REST clients (consistent & hardened)
– Controller is single point of API integration (versus tens of boxes)
Benefits
• No DevOps cost for network automation
• Print REST from CLI/GUI (accelerate DevOps through NetOps)
• Scalable M2M API interaction
dt-controller1# debug rest
***** Enabled display rest mode *****
dt-controller1# show tenant blue
REST-POST: POST http://127.0.0.1:8080/api/v1/data/controller/core/aaa/audit-event {"attribute": [{"value": ”show tenant blue", "key": "cmd_args"}], "event-type": "cli.command", "session-cookie": "yx6pjq6cwo5YXZwHsDyw6Z_3Zm5PITwE”}
REST-POST: http://127.0.0.1:8080/api/v1/data/controller/core/aaa/audit-event done 0:00:00.003089
40
Internal Use - Confidential
Big Cloud Fabric is open, simple and economical alternative to ACIA Better Alternative to Cisco ACI
Challenges:
• Proprietary: HW-heavy, Vendor lock-in
• Complex: Box-by-box control plane, Pro-service necessary,
Confusing Taxonomy
• Cost: Still Expensive
• Company: Reorgs, Talent depletion, Low Morale
Confusing & Unnecessary Taxonomy:
• Promise theory, Contracts, EPG
Advantages:
• Open: HW/SW disaggregation, Vendor choice
• Simple: No complex protocols, zero-touch
• Economical: Commodity Bare-metal HW
APIC: Application Policy Infrastructure Controller
Very large System Integrator Feedback
• “Four people worked on different tests but every one of us
was impressed with the ease of use and quality”
• “Done demos internally and to several customers…
consistently received very positive feedback, even from SDN
skeptics”
• “ACI is complicated, customers are confused even when
Cisco presents it, has only partial CLI”
APIC is a policy controller, not SDN controller
Gartner: Cisco ACI is NOT SDN