vmworld 2013: vsphere distributed switch – design and best practices
DESCRIPTION
VMworld 2013 Vyenkatesh (Venky) Deshpande, VMware Marcos Hernandez, VMware Learn more about VMworld and register at http://www.vmworld.com/index.jspa?src=socmed-vmworld-slideshareTRANSCRIPT
vSphere Distributed Switch –
Design and Best Practices
Vyenkatesh (Venky) Deshpande, VMware
Marcos Hernandez, VMware
NET5521
#NET5521
2
Session Objective
New capabilities in VDS
VDS can meet your design requirements
Provide Common best practices while designing with VDS
3
Recommended Sessions & Labs
VSVC4966 – vSphere Distributed Switch – Technical Deep Dive
VSVC5103 - vSphere Networking and vCloud Networking Suite
Best Practices and Troubleshooting
You can check out VSS to VDS Migration workflow and new VDS
features in the lab HOL-SDC-1302
NET5266 - Bringing Network Virtualization to VMware environments
with NSX
NET5654 - Troubleshooting VXLAN and Network Services in a
Virtualized Environment
4
Agenda
Overview of VDS and New Features in 5.5
Common Customer Deployments
Design and Best Practices
NSX and VDS
5
VDS Overview and 5.5 Features
6
vSphere Distributed Switch (VDS)
vSphere Distributed Switch
Manage a Datacenter wide switch vs. Individual switches per host
Advanced feature support
Higher Scale
Foundation for your Network Virtualization Journey
7
vSphere Distributed Switch (VDS) Architecture
vSphere vSphere
vSphere Distributed Switch
Host 1 Host 2
Legend :
dvPG-A
dvPG-B
Data Plane Data Plane
Data Plane : Handles the packet switching function
VMware vCenter Server
Management Plane
vSphere Distributed Switch
Management Plane : Allows to configure various parameters of the distributed switch
vmnic0 vmnic1 vmnic0 vmnic1
dvUplink PG
dvUplink
dvuplink1 dvuplink2
8
VDS Enhancements in vSphere 5.5
Visibility & Troubleshooting
Performance and Scale
Host Level Packet Capture
Tool (tcpdump). Available
for Standard Switch as well
Enhanced LACP
Enhanced SR-IOV
40 Gig NIC support
Packet Classification
Traffic Filtering (ACLs)
DSCP Marking (QoS)
vSphere Distributed Switch
9
LACP Enhancements
vSphere
vSphere Distributed Switch
Host
Physical switches
LACP
Communication
Link Aggregation Control
Protocol
Standards based – 802.3ad
Automatic negotiation of link aggregation
parameters
Advantages
Aggregates link BW and provides
redundancy
Detects link failures and cabling mistakes
and automatically reconfigures
Enhancements
Support for 64 LAGs per VDS and per
Host
Support for 22 different hashing
algorithms
10
Common Customer Deployments
11
VDS in the Enterprise
VMware vCentServer vCenter Server
Multiple VDS per VC (128)
VDS can span multiple Clusters
Hundreds of Hosts per VDS
Central Management for DC and
ROBO environments
Role Based management control
VDS VDS
ROBO 1 ROBO 2
VDS VDS VDS
Cluster 1 Cluster 2 Cluster 3 Cluster 4
Data
Cente
r
12
Design Best Practices
13
Infrastructure Design Goals
Reliable
Secure
Performance
Scalable
Operational
14
Infrastructure Types Influence Your Design Decisions
Available Infrastructure
• Type of Servers
• Type of Physical Switches
Servers
• Rack mount or Blade
• Number of Ports and Speed. For example, Multiple 1 Gig or 2 – 10 Gig
Physical Switches
• Managed and un-managed
• Protocol and features support
Example Deployment – 2 – 10 Gig Server configuration
15
Reliable - Connectivity
16
Physical Connection Options
vSphere VDS
vSphere
VDS
vSphere
VDS
vSphere
VDS
Port Group – Teaming
Port ID, MAC Hash,
Explicit Failover, LBT
One Physical Switch Two Physical Switches One Physical Switch
with Ether Channel Two Physical Switches
in MLAG configuration
Port Group – Teaming
IP Hash Port Group – Teaming
LACP
Port Group – Teaming
Port ID, MAC Hash,
Explicit Failover, LBT
MLAG/vPC
17
Connectivity Best Practices
Avoid Single point of Failure
• Connect two or more physical NICs to a VDS
• Preferably connect those physical NICs to separate physical switches
Configure Port groups with appropriate teaming setting based on
the physical switch connectivity and configuration. For example
• Use IP hash when Ether channel is configured on Physical Switch
Configure Port Fast and BPDU guard on Access Switch Ports
• No STP running on virtual switches
• No loop created by virtual switch
Trunk all Port group VLANs on Access Switch ports
18
Spanning Tree Protocol Boundary
vSphere vSphere
vSphere Distributed Switch
Switch Port
Configuration: Port Fast
BPDU Guard
VLAN 10,20
Switch Port
Configuration: Port Fast
BPDU Guard
VLAN 10.20
Physical Network
Virtual Network
Spanning Tree Protocol Boundary
No Spanning Tree
Support
No BPDU
generated
19
Teaming Best Practices
Link Aggregation mechanisms do not double the BW
• Hashing algorithm performs better in some scenarios. For example
• Web servers accessed by different users have enough variation in IP Src and Dest
fields and can utilize links effectively
• However, few workloads accessing a NAS array doesn’t have any variation in
the packet header fields. Traffic might end up on only one physical NIC
Why Load Based Teaming is better ?
Takes into account link utilization
Checks Utilization of Links every 30 seconds
No special configuration required on the physical switches
20
Load Based Teaming
1 2 3 4
10 11
VM1 VM2
vMotion
1 2 3 4
10 11
VM2
Network Traffic Bandwidth
vMotion traffic 7 Gig
VM1 traffic 5 Gig
VM2 traffic 2 Gig
10 Gig 2 Gig 7 Gig 7 Gig
VDS VDS
VM1
vMotion
Rebalance
21
Security/Isolation
22
Traffic Types Running on a vSphere Host
vSphere
PG-A PG-B
Host
VDS
PG-C PG-E PG-D
Mgmt Traffic
vmk3
vMotion
Traffic
vmk4
FT
Traffic
vmk2
NFS
Traffic
vmk1
10 Gig 10 Gig
23
Security Best Practices
Provide Traffic Isolation using VLANs
• Each Port group can be associated with different VLAN
Keep default Security settings on the Port group
• Promiscuous Mode – Reject
• MAC address Changes – Reject
• Forged Transmit – Reject
While utilizing PVLAN feature make sure Physical Switches are
also configured with Primary, Secondary VLAN configuration
Enable BPDU filter property at Host level to prevent DoS attack
situation due to compromised virtual machines
Make use Access Control List Feature (5.5)
24
Performance
25
Why Should You Care About Performance?
As more workloads are getting virtualized, 10 Gig pipes
are getting filled
Some workloads have specific BW and latency requirements
• Business Critical applications
• VOIP applications
• VDI application
Noisy Neighbors problem has to be addressed
• vMotion is very BW intensive and can impact other traffic types
• General Purpose VM traffic can impact other critical applications such
as VOIP application
26
Administrator
Mgmt vMotion
Teaming Policy vSphere Distributed Switch
vSphere Distributed
Port groups
Network I/O Control
VM
Traffic
Scheduler
Shaper
Scheduler
Shaper
FT NFS
Traffic Shares Limit
(Mbps)
802.1p
VM Traffic 30 - 4
vMotion 20 - 3
Mgmt 5 - 7
FT 10 - 6
NFS 20 - 5
Port 1
Port 2
10 Gig 10 Gig
Infrastructure Traffics
4000
Limits
Host
Shares %
BW
Link BW
10 Gig
30 30/50 3/5*10 = 6
20 20/50 2/5*10 = 4
Total 50
27
Administrator
Mgmt vMotion
Teaming Policy vSphere Distributed Switch
vSphere Distributed Port groups
Business Critical Applications and User Defined Traffic Types
VM
Traffic
Scheduler
Shaper
Scheduler
FT NFS
Traffic Shares Limit
(Mbps)
802.1p
App1 10 - 7
App2 10 - 6
VM Traffic 10 - 4
vMotion 20 - 3
Mgmt 5 - 7
FT 10 - 6
NFS 20 - 5
Port 1
Port 2
10 Gig 10 Gig
App 2
Traffic App 1
Traffic
Shaper
Host
28
End to End QoS
How to make sure that the Application traffic flowing through
Physical Network Infrastructure is also Prioritized ?
Two types of Tagging or Marking supported
• COS – Layer 2 Tag
• DSCP Marking – Layer 3 Tag
0x8100 COS VLAN D
16 bits 3 bits 12 bits 1 bit
802.1Q Header
DSCP ECN
6 bits 2 bits
Version H Length TOS/DS P Length …..
IP Header
29
Tagging at Different Level
vSphere vSphere Switch
Physical
Network
DSCP
COS
vSphere vSphere Switch
Physical
Network
DSCP
COS
vSphere vSphere Switch
Physical
Network
DSCP
COS
Guest Tagging Virtual Switch Tagging Physical Switch Tagging
VDS can pass VM QoS
markings downstream
NIOC can’t assign
separate queue based
on the tag
Admins lose control
VDS implements 802.1p and/or
DSCP marking
Preferred option
Single Edge QoS enforcement
point
QoS marking or remarking
done in the physical switch
and/or router
Burdensome QoS management
on each edge device (e.g. ToR)
30
Congestion Scenario in the Physical Network
vSphere vSphere Switch
vSphere vSphere Switch
Higher Tagged Traffic
Un Tagged Traffic
Lower Tagged Traffic
Congested Switch
Physical Network
31
Mgmt vMotion
Per Port Traffic Shaping
VM
Traffic
10 Gig 10 Gig
Ingress Egress
Time
BW
Average BW
Peak BW
Burst Size
Ingress and Egress
Parameters
Average Bandwidth
Kbps
Peak Bandwidth
Kbps
Burst Size
Kbytes
Token
Bucket
32
Other Performance Related Decisions
Need more BW for Storage
• If iSCSI, utilize Multi-Pathing.
• MTU configuration – Jumbo frame
• LBT can’t work for iSCSI traffic because of port binding requirements
Need more BW for vMotion
• Use Multi-NIC vMotion.
• LBT doesn’t split the vMotion traffic to multiple Physical NICs.
Latency Sensitive application – Care about Micro seconds
• Utilize SR-IOV
• Doesn’t support vMotion, HA and DRS features
33
Scalable
34
Scale
Scaling Compute Infrastructure
Adding Hosts to Clusters
Adding new Clusters
Impact on VDS Design
VDS can span across 500 hosts
VDS
Cluster 1 Cluster 2 Cluster 3 Cluster 4
Data
Cente
r
VDS
Cluster 1 Cluster 2 Cluster 3 Cluster 4
Data
Cente
r
Scaling number of users or
applications
More Virtual Machines connected to
isolated networks (VLANs)
Impact on VDS Design
Separate port groups for each application
– 10,000 port groups support
Number of virtual ports - 60,000
Dynamic Port management (Static Ports)
35
Operational
36
How to Operate Your Virtual Network?
Major concerns
• Lost visibility into traffic from VM to VM on the same Host
• How do I troubleshoot configuration issues?
• How do I troubleshoot connectivity issues?
Make use of VDS features
• Netflow and Port Mirroring
• Network Health Check detects mis-configuration across virtual
and physical switches
• Host level Packet Capture allows you to monitor traffic at vnic,
vmknic and vmnic level
37
NSX and VDS
38
VMware NSX Functional System Overview
vSphere vSphere vSphere vSphere
vSwitch vSwitch vSwitch vSwitch
Hosts
Data Plane
Operations
UI
Logs/Stats
CMP
Consumption
Tenant UI
API
Control Plane Run-time state
Management Plane API
API, config, etc.
HA, scale-out
NSX Manager
NSX Controller
vCenter Server
39
VXLAN Protocol Overview
Ethernet in IP overlay network
Entire L2 frame encapsulated in
UDP
50+ bytes of overhead
Decouples Physical network
from the Logical
24 bits VXLAN ID identifies 16 M
Logical networks
VMs do NOT see VXLAN ID
Physical Network devices don’t see
VMs MAC and IP address
VTEP (VXLAN Tunnel End
Point)
VMkernel interface which serves as
the endpoint for encapsulation/de-
encapsulation of VXLAN traffic
VXLAN can cross Layer 3
network boundaries
Technology submitted to IETF
for standardization
• With Cisco, Citrix, Red Hat,
Broadcom, Arista and Others
40
VXLAN Configuration on VDS
vSphere Host
VM1
VXLAN Transport Network
vSphere Host
VM2
vSphere Host
VXLAN 5001
VTEP1 10.20.10.10 VTEP2 10.20.10.11 VTEP3 10.20.11.10
vSphere Host
VTEP4 10.20.11.11
VM3 VM4
VXLAN Transport Subnet A 10.20.10.0/24 VXLAN Transport Subnet B 10.20.11.0/24
vSphere Distributed Switch
41
For More Details on VXLAN attend NET5654 - Troubleshooting VXLAN and Network
Services in a Virtualized Environment
42
Key Takeaways
VDS is flexible and scalable to meet your design requirements.
VDS simplifies the deployment and operational aspects
of virtual network
Make use of NIOC and LBT feature to improve utilization
of your I/O resources
VDS is a key component of NSX Platform
43
Q&A
Paper: http://www.vmware.com/resources/techresources/10250
http://blogs.vmware.com/vsphere/networking
@VMWNetworking
44
Other VMware Activities Related to This Session
HOL:
HOL-SDC-1302
vSphere Distributed Switch from A to Z
Group Discussions:
NET1000-GD
vSphere Distributed Switch with Vyenkatesh Deshpande
THANK YOU
vSphere Distributed Switch –
Design and Best Practices
Vyenkatesh Deshpande, VMware
Marcos Hernandez, VMware
NET5521
#NET5521
48
Backup: Example Design
49
VDS in Rack Server Deployment: Two 10 Gig Ports
Access
Layer
Aggregation
Layer
. . . . . . . . . . . . . . . . .
Cluster 1 Cluster 2
ESXi ESXi ESXi ESXi
vSphere Distributed Switch
Legend :
PG-A
PG-B L2 Switch
Router
50
Option1: Static Design – Port Group to NIC Mapping
Traffic Type Port
Group
Teaming
Option
Active
Uplink
Standby
Uplink
Unused
Uplink
Virtual Machine PG-A LBT dvuplink1/
dvuplink2 None None
NFS PG-B Explicit
Failover dvuplink1 dvuplink2 None
FT PG-C
Explicit
Failover
dvuplink2 dvuplink1 None
Management PG-D
Explicit
Failover
dvuplink2 dvuplink1 None
vMotion PG-E
Explicit
Failover
dvuplink2 dvuplink1 None
51
Option2: Dynamic Design – Use NIOC and Configure Shares and Limits
Need Bandwidth information for different traffic types
• NetFlow
Bandwidth Assumption
• Management – Less than 1 Gig
• vMotion – 2 Gig
• NFS – 2 Gig
• FT – 1 Gig
• Virtual Machine – 2 Gig
Shares calculation
• Equal shares to vMotion, NFS and Virtual Machine
• Lower shares to Management and FT
52
Option2: Dynamic Design – Use NIOC and Configure Shares and Limits
Traffic
Type
Port
Group
Teaming
Option
Active
Uplink
Standby
Uplink
NIOC
Shares
NIOC
Limits
Virtual
Machine PG-A LBT dvuplink1,2 None 20 -
NFS PG-B LBT dvuplink1,2 None 20 -
FT PG-C LBT
dvuplink1,2 None 10 -
Mgmt. PG-D LBT
dvuplink1,2 None 5 -
vMotion PG-E LBT dvuplink1,2 None 20 -
53
Dynamic Design Option with NIOC and LBT – Pros and Cons
Pros
• Better utilized I/O resources through traffic management
• Logical separation of traffic through VLAN
• Traffic SLA maintained through NIOC shares
• Resiliency through Active-Active Paths
Cons
• Dynamic traffic movement across physical infrastructure need all paths
to be available and handle any traffic characteristics.
• VLAN expertise