QOS Openvswitch - en

Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

Open vSwitch QoS

Netlter Workshop, Seville, Spain

Simon Horman <[email protected]> Horms Solutions Ltd., Tokyo

October 2010

Open vSwitch

Server
Open vSwitch Datapath Open vSwitch Controller

Flexibility for Networking in Virtualised Environments Flexible Controller in User-Space Fast Datapath in Kernel

Open vSwitch Availability

Available from openvswitch.org Development code is available in git Announce, discussion and development mailing lists User-space (controller and tools) is under the Apache license Kernel (datapath) is under the GPLv2 Shared headers are dual-licensed

Open vSwitch Concepts

A switch contains ports A port may have one or more interfaces


Bonding allows more than once interface per port

Packets are forward by ow

Packets are Managed as Flows

A ow may be identied by any combination of


Input port VLAN ID (802.1Q) Ethernet Source MAC address Ethernet Destination MAC address IP Source MAC address IP Destination MAC address TCP/UDP/... Source Port TCP/UDP/... Destination Port

Packets are Managed as Flows


1 2

The rst packet of a ow is sent to the controller The controller programs the datapaths actions for a ow
Usually one, but may be a list Actions include:
Forward to a port or ports, mirror Encapsulate and forward to controller Drop

3 4

And returns the packet to the datapath Subsequent packets are handled directly by the datapath
Server
Open vSwitch Datapath
1 2 4 3

Open vSwitch Controller

Network Scalability Problems in Virtualised Environments

Migration VLANs QoS Management

Migration

KVM and Xen provide Live Migration With bridging, IP address migration must occur with in the same L2 network Open vSwitch avoids this problem using GRE tunnels

VLANs

Per-Customer VLANs are desirable for security reasons But there is a limit of 4094 VLANs

More VLANs

Two, apparently competing, approaches


1

IETF / Cisco
RFC5517 Private VLANs

IEEE
802.1ad Provider Bridges (Q-in-Q) 802.1ah Provider Backbone Brides (MAC-in-MAC)

RFC5517 Private VLANs

Uses existing 802.1Q framing


Simple to implement (in software/rmware)

Makes use of pairs of VIDs


Requires all switches to support of Private VLANs otherwise switch tables may not merge

Provides L2 broadcast isolation


Forwarding may occur at L3 Requires the router to perform proxy ARP

Currently not supported by Open vSwitch

RFC5517 Private VLANs

Three VLAN classications Promiscuous


May communicate with endpoints on any port e.g.: Gateway, Management Host

Community
May only communicate with endpoints on promiscuous ports or ports belonging to the same comunity e.g.: Dierent hosts belonging to the same customer

Isolated
May only communicate with endpoints on promiscuous ports e.g.: Hosts that only require access to the gateway

Private VLANs Domain View

P C1 C2 I

Promiscous domain (P)


May communicate with endpoints in the same domain and sub-domains

Two community sub-domains (C1 , C2 )


May communicate with endpoints in the same domain and parent-domain

Isolated sub-domain (I)


May communicate with endpoints in the parent domain May not communicate with endpoints in the same domain

802.1ad Provider Bridges (Q-in-Q)

Current standard is 802.1ad-2005, Approved December 2005 Builds on 802.1Q New Framing
C-VID (inner)
Renamed 802.1Q VID There may be more than one C-VID (inner-inner, ...)

S-VID (outer)
Dierent ether-type to C-VID May be translated

Currently not supported by Linux Kernel / Open vSwitch

802.1ad Framing Provider Bridges


DA SA S-VID C-VID VID Destination MAC address Source MAC addresss Service VLAN ID Customer VLAN ID VLAN ID

DA SA Payload

DA SA VID Payload

DA SA S-VID C-VID Payload

802.1 802.1Q

802.1ad

802.1ah Provider Backbone Bridges (MAC-in-MAC)

Current standard is 802.1ah-2008, Approved August 2008 Builds on 802.1ad New Framing
MAC encapsulation provides full Client VLAN isolation
Inner MAC is unknown outside of its scope

I-SID: Up to 224 16million backbone services I-VID semantics are the same as the S-VLAN
Only edge switches need to be Provider Backbone Bridge aware Core switches need only be Provider Bridge (802.1ad) aware

Currently not supported by Linux Kernel / Open vSwitch

802.1ah Framing Provider Backbone Bridges


B-DA Backbone Destination MAC address B-SA Backbone Source MAC address B-VID Backbone VLAN ID I-SID DA SA S-VID C-VID VID DA SA Payload Payload Service ID Destination MAC address Source MAC addresss Service VLAN ID Customer VLAN ID VLAN ID DA SA VID DA SA S-VID C-VID Payload

B-DA B-SA B-VID I-SID DA SA S-VID C-VID Payload

802.1 802.1Q

802.1ad

802.1ah

QoS

Open vSwitch QoS capabilities


1 2

Interface rate limiting Port QoS policy

QoS: Interface rate limiting

A rate and burst can be assigned to an Interface Conceptually similar to Xens netback credit scheduler # ovs-vsctl set Interface tap0 ingress_policing_rate=100000 # ovs-vsctl set Interface tap0 ingress_policing_burst=10000 Simple Appears to work as expected

QoS: No interface rate limiting example

# netperf -4 -t UDP_STREAM -H 172.17.50.253 -- -m 8972 UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_ to +172.17.50.253 (172.17.50.253) port 0 AF_INET Socket Message Elapsed Messages Size Size Time Okay Errors Throughput bytes bytes secs # # 10^6bits/sec 120832 109568 8972 10.01 10.01 146797 146620 0 1052.60 1051.33

tap networking used jumbo frames required to reach line speed (210Mbits/s with 1500 byte frames) virtio should do better?

QoS: Interface rate limiting example

# netperf -4 -t UDP_STREAM -H 172.17.50.253 UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_ to 172.17.50.253 (172.17.50.253) port 0 AF_INET Socket Message Elapsed Messages Size Size Time Okay Errors Throughput bytes bytes secs # # 10^6bits/sec 120832 109568 8972 10.01 10.01 149735 14684 0 1073.66 105.29

Dierence in sent and received packets indicates a ow control problem. virtio should do better?

QoS: Port QoS policy

A port may be assigned one ore more QoS policy Each QoS policy consists of a class and a qdisc
Classes and qdisc use the Linux kernels tc implementation Only HTB classes are supported at this time Each class has a single qdisc associated with it The class of a ow is chosen by the controller

The QoS policy (i.e. class) of a ow is chosen by the controller

QoS: Port QoS policy example


Programming the Datapath 1:# 2: 3: 4: 5: 6: 7: 8: 9: ovs-vsctl set port eth1 qos=@newqos \ -- --id=@newqos create qos type=linux-htb \ other-config:max-rate=200000000 queues=0=@q0,1=@q1 \ -- --id=@q0 create queue \ other-config:min-rate=100000000 \ other-config:max-rate=100000000 \ -- --id=@q1 create queue \ other-config:min-rate=50000000 \ other-config:max-rate=50000000

QoS: Port QoS policy example

Hard-coding the controller # ovs-ofctl add-flow br0 "in_port=2 ip nw_dst=172.17.50.253 \ idle_timeout=0 actions=enqueue:1:0" # ovs-ofctl add-flow br0 "in_port=3 ip nw_dst=172.17.50.253 \ idle_timeout=0 actions=enqueue:1:1" Only suitable for testing

QoS: Port QoS policy example


Guest 0: # netperf -4 -t TCP_STREAM -H 172.17.50.253 -l 30 -- -m 8972 TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 172.17.50.253 (172.17.50.253) port 0 AF_INET Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 87380 16384 8972 30.01 99.12 Guest 1: # netperf -4 -t TCP_STREAM -H 172.17.50.253 -l 30 -- -m 8972 ... 87380 16384 8972 30.14 49.56

QoS: Port QoS policy controller improvements

Add a default queue to the Port table Add enqueue to the FLOOD and NORMAL ports or use NOX (a dierent controller)

Conclusion

Open vSwitch is aimed at addressing short-comings in using bridging in virtualised environments It is a young project and there is much scope to contribute
Extended VLAN support
Private VLANs 802.1ad 802.1ah

Improved QoS
Add a default queue to the Port table Add enqueue to the FLOOD and NORMAL ports or use NOX (a dierent controller)

High-Level Management

You might also like