Internet Engineering Task Force L. Kreeger
Internet-Draft D. Dutt
Intended status: Informational Cisco
Expires: August 2, 2012 T. Narten
IBM
D. Black
EMC
M. Sridharan
Microsoft
January 30, 2012
Network Virtualization Overlay Control Protocol Requirements
draft-kreeger-nvo3-overlay-cp-00
Abstract
The document draft-narten-nvo3-overlay-problem-statement-01 discusses
the needs for network virtualization using overlay networks in highly
virtualized data centers. The problem statement outlines a need for
control protocols to facilitate running these overlay networks. This
document outlines the high level requirements to be fulfilled by the
control protocols.
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on August 2, 2012.
Copyright Notice
Copyright (c) 2012 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
Kreeger, et al. Expires August 2, 2012 [Page 1]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. Control Plane Protocol Functionality . . . . . . . . . . . . . 5
3.1. Inner to Outer Address Mapping . . . . . . . . . . . . . . 8
3.2. Underlying Network Multi-Destination Delivery
Address(es) . . . . . . . . . . . . . . . . . . . . . . . 8
3.3. VN Connect/Disconnect Notification . . . . . . . . . . . . 9
3.4. VN Name to VN-ID Mapping . . . . . . . . . . . . . . . . . 10
4. Control Plane Characteristics . . . . . . . . . . . . . . . . 10
5. Security Considerations . . . . . . . . . . . . . . . . . . . 12
6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 12
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 13
Kreeger, et al. Expires August 2, 2012 [Page 2]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
1. Introduction
The document draft-narten-nvo3-overlay-problem-statement-01 discusses
the needs for network virtualization using overlay networks in highly
virtualized data centers. It focuses the problem less on the
particular encapsulation, or even what address families are carried
inside/outside the overlay, but instead on the control protocol
issues that need to be addressed in order to provide a solution. The
problem statement discusses the use of virtual network overlays where
the encapsulation/decapsulation is performed by the first hop switch
in the data center, which could be either a virtual switch residing
in the hypervisor, or a physical access switch connected to a server
or Network Service Appliance.
2. Terminology
This document uses the following terminology:
VN: Virtual Network. This is one instance of a virtual overlay
network. Two Virtual Networks are isolated from one another and
may use overlapping addresses.
VN-ID: Virtual Network Identifier. This is the ID value that is
carried in each data packet in the overlay encapsulation that
identifies the Virtual Network the packet belongs to. It should
be a large enough ID space to not be a limiting factor within an
administrative domain managing the ID space. There are several
technologies which encapsulate using a 24 bit ID value, e.g. PBB,
SPBM, LISP, OTV, TRILL Fine-grained labels, VXLAN, NVGRE.
OBP: Overlay Boundary Point. This is a network entity that is on
the edge boundary of the overlay. It performs encapsulation to
send packets to other OBPs across an Underlying Network for
decapsulation. An OBP could be implemented as part of a virtual
switch within a hypervisor, a physical switch or router, a Network
Service Appliance or even be embedded within an End Station.
Underlying Network: This is the network that provides the
connectivity between the OBPs. The Underlying Network can be
completely unaware of the VN of packets carried within the
encapsulation. Addresses within the Underlying Network are also
referred to as "outer addresses" because they exist in the outer
encapsulation. The Underlying Network can use a completely
different protocol (and address family) from that of the overlay.
Kreeger, et al. Expires August 2, 2012 [Page 3]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
Data Center: A physical complex housing physical servers, network
switches and routers, Network Service Appliances and networked
storage. The purpose of a Data Center is to provide application
and/or compute and/or storage services. One such service is
virtualized data center services, also known as Infrastructure as
a Service.
Network Service Appliance: A stand-alone physical device or a
virtual device that provides a network service, such as a
firewall, load balancer, etc. Such appliances may embed OBP
functionality within them in order to more efficiently operate as
part of a virtualized network.
VM: Virtual Machine. Several Virtual Machines can share the
resources of a single physical computer server using the services
of a Hypervisor (see below definition).
Hypervisor: Server virtualization software running on a physical
compute server that hosts Virtual Machines. The hypervisor
provides shared compute/memory/storage and network connectivity to
the VMs that it hosts. Hypervisors often embed a Virtual Switch
(see below).
Virtual Switch: A function within a Hypervisor (typically
implemented in software), that provides similar services to a
physical Ethernet switch. It switches Ethernet frames between
VMs' virtual NICs within the same physical server, or between a VM
and a physical NIC card connecting the server to a physical
Ethernet switch. It also enforces network isolation between VMs
that should not communicate with each other.
End Station: This is an end device which connects to a VN. The End
Station is unaware of how the VN is implemented. OBPs
encapsulate/decapsulate on the behalf of these End Stations. An
End Station can be a VM, a physical server, or a Network Service
Appliance. End Station addresses are also referred to as "inner
addresses" because they exist inside of the overlay encapsulation
payload.
Tenant: A customer who consumes virtualized data center services
offered by a cloud service provider. A single tenant may consume
one or more Virtual Data Centers hosted by the same cloud service
provider.
VDC: Virtual Data Center. A container for virtualized compute,
storage and network services. Managed by a single tenant, a VDC
can contain multiple VNs and multiple End Stations that are
connected to one or more of these VNs.
Kreeger, et al. Expires August 2, 2012 [Page 4]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
VN Name: A globally unique name for a VN. The VN Name is not
carried in data packets originating from End Stations, but must be
mapped into an appropriate VN-ID for a particular encapsulating
technology. Using VN Names rather than VN-IDs to identify VNs in
configuration files and control protocols increases the
portability of a VDC and its associated VNs when moving among
different administrative domains (e.g. switching to a different
cloud service provider).
3. Control Plane Protocol Functionality
The problem statement
(draft-narten-nvo3-overlay-problem-statement-01), discusses the needs
for a control plane protocol (or protocols) to populate each OBP with
the state needed to peform its functions.
Note that an OBP may provide overlay encapsulation/decapsulation
packet forwarding services to End Stations that are co-resident
within the same device (e.g. when the OBP is embedded within a
hypervisor or a Network Service Appliance), or to End Stations that
are externally connected to the OBP (e.g. a physical Network Service
Appliance connected to an access switch containing the OBP).
The following figures show examples of scenarios in which the OBP is
co-resident within the same device as the End Stations connected to a
given VN, and when the OBP is externally located within the access
switch.
Hypervisor
+-----------------------+
| +--+ +-------+---+ |
| |VM|---| | | |
| +--+ |Virtual|OBP|----- Underlying
| +--+ |Switch | | | Network
| |VM|---| | | |
| +--+ +-------+---+ |
+-----------------------+
Hypervisor with an Embedded OBP
Figure 1
Kreeger, et al. Expires August 2, 2012 [Page 5]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
Hypervisor Access Switch
+------------------+ +-----+-------+
| +--+ +-------+ | | | |
| |VM|---| | | VLAN | | |
| +--+ |Virtual|---------+ OBP | +--- Underlying
| +--+ |Switch | | Trunk | | | Network
| |VM|---| | | | | |
| +--+ +-------+ | | | |
+------------------+ +-----+-------+
Hypervisor with an External OBP
Figure 2
Network Service Appliance
+---------------------------+
| +------------+ +-----+ |
| |Net Service |---| | |
| |Instance | | | |
| +------------+ | OBP |------ Underlying
| +------------+ | | | Network
| |Net Service |---| | |
| |Instance | | | |
| +------------+ +-----+ |
+---------------------------+
Network Service Appliance (physical or virtual) with an Embedded OBP
Figure 3
Kreeger, et al. Expires August 2, 2012 [Page 6]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
Network Service Appliance Access Switch
+--------------------------+ +-----+-------+
| +------------+ |\ | | | |
| |Net Service |----| \ | | | |
| |Instance | | \ | VLAN | | |
| +------------+ | |---------+ OBP | +--- Underlying
| +------------+ | | | Trunk| | | Network
| |Net Service |----| / | | | |
| |Instance | | / | | | |
| +------------+ |/ | | | |
+--------------------------+ +-----+-------+
Physical Network Service Appliance with an External OBP
Figure 4
In the examples above where the OBP functionality is located in the
physical access switch, the physical VLAN Trunk connecting the
Hypervisor or Network Services Appliance to the external OBP only
needs to carry locally significant (e.g. link local) VLAN tag values.
These tags are only used to differentiate two different VNs as
packets cross the wire to the external OBP. When the OBP receives
packets, it uses the VLAN tag to identify the VN the End Station
belongs to, strips the tag, and adds the appropriate overlay
encapsulation for that VN.
Given the above, a control plane protocol is necessary to provide an
OBP with the information it needs to maintain its own internal state
necessary to carry out its forwarding functions as explained in
detail below.
1. An OBP maintains a per-VN table of mappings from End Station
(inner) addresses to Underlying Network (outer) addresses of
remote OBPs.
2. An OBP maintains per-VN state for delivering multicast and
broadcast packets to other End Stations. Such state could
include a list of multicast addresses and/or unicast addresses on
the Underlying Network for the OBPs associated with a particular
VN.
3. Devices (such as a Hypervisor or Network Service Appliance)
utilizing an external OBP need to "attach to" and "detach from"
an OBP. Specifically, they will need a protocol that runs across
the L2 link between the two devices that identifies the End
Station and VN Name for which the OBP is providing service. In
addition, such a protocol will identify a locally significant tag
Kreeger, et al. Expires August 2, 2012 [Page 7]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
(e.g., an 802.1Q VLAN tag) that can be used to identify the data
frames that flow between the End Station and VN.
4. An OBP needs a mapping from each unique VN name to the VN-ID
value used within encapsulated data packets within the
administrative domain that the VN is instantiated.
3.1. Inner to Outer Address Mapping
When presented with a data packet to forward to an End Station within
a VN, the OBP needs to know the mapping of the End Station
destination (inner) address to the (outer) address on the Underlying
Network of the remote OBP which can deliver the packet to the
destination End Station.
A protocol is needed to provide this inner to outer mapping to each
OBP that requires it and keep the mapping updated in a timely manner.
Timely updates are important for maintaining connectivity between End
Stations when one End Station is a VM
Note that one technique that could be used to create this mapping
without the need for a control protocol is via data plane learning;
However, the learning approach requires packets to be flooded to all
OBPs participating in the VN when no mapping exists. One goal of
using a control protocol is to eliminate this flooding.
3.2. Underlying Network Multi-Destination Delivery Address(es)
Each OBP needs a way to deliver multi-destination packets (i.e.
broadcast/multicast) within a given VN to each remote OBP which has a
destination End Station for these packets. Three possible ways of
accomplishing this:
o Use the multicast capabilities of the Underlying Network.
o Have each OBP replicate the packets and send a copy across the
Underlying Network to each remote OBP currently participating in
the VN.
o Use one or more distribution servers which replicates the packets
on the behalf of the OBPs.
Whichever method is used, a protocol is needed to provide on a per VN
basis, one or more multicast address (assuming the Underlying Network
supports multicast), and/or one or more unicast addresses of either
the remote OBPs which are not multicast reachable, or of one or more
distribution servers for the VN.
Kreeger, et al. Expires August 2, 2012 [Page 8]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
The protocol must also keep the list of addresses up to date in a
timely manner if the set of OBPs for a given VN changes over time.
For example, the set of OBPs for a VN could change as VMs power on/
off or migrate to different hypervisors.
3.3. VN Connect/Disconnect Notification
As the previous figures illustrated, OBPs may be embedded within a
device (such as a Hypervisor or Network Service Appliance), or within
an external networking device (e.g. an access switch). Using an
external network device as the OBP can provide an offload of the
encapsulation / decapsulation function and the protocol overheads
which may provide performance improvements and/or resource savings to
the client device making use of the external OBP.
When an OBP is external, a protocol is needed between a client device
making use of the external OBP and the OBP itself in order to make
the OBP aware of the changing VN membership requirements of the
client device. A key driver for using a protocol rather than using
static configuration of the exernal OBP is because the VN
connectivity requirements can change frequently as VMs are brought
up, moved and brought down on various hypervisors throughout the data
center.
The OBP must be notified when a client device requires connection to
a particular VN and when it no longer requires connection. This
protocol should also provide the inner End Station addresses within
the VN that the client device contains (e.g. the virtual MAC address
of a VMs virtual NIC) to the external OBP. In addition, the external
OBP must provide a local tag value for each connected VN to the
client device to use for exchange of packets between the client
device to the OBP (e.g. a locally significant 802.1Q tag value).
The Identification of the VN in this protocol should preferably be
made using a globally unique VN Name. A globally unique VN Name
facilitates portability of a Tenant's Virtual Data Center. When a VN
within a VDC is instantiated within a particular administrative
domain, it can be allocated a VN-ID which only the OBP needs to use.
A client device that is making use of an offloaded OBP only needs to
communicate the VN Name to the OBP, and get back a locally
significant tag value. Ideally the VN Name should be compact as well
unique to minimize protocol overhead. Using a Universally Unique
Identifier (UUID) as discussed in RFC 4122, would work well because
it is both compact and a fixed size and can be generated locally with
a high likelihood of being unique.
Kreeger, et al. Expires August 2, 2012 [Page 9]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
3.4. VN Name to VN-ID Mapping
Once an OBP (embedded or external) receives a VN connect indication
with a specified VN name, the OBP must find the VN-ID value to
encapsulate packets with. The OBP serving that hypervisor needs a
way to get a VN-ID allocated or receive the already allocated VN-ID
for a given VN Name. A protocol for an OBP to get this mapping may
be a useful function.
4. Control Plane Characteristics
OBPs are expected to be implemented within hypervisors or access
switches, or even within a Network Service Appliance. Any resources
used by these protocols (e.g. processing or memory) takes away
resources that could be better used by these devices to perform their
intended functions (e.g. providing resoures for hosted VMs).
A large scale data center may contain hundreds of thousands of these
OBPs (which may be several independent implementations); Therefore,
any savings in per-OBP resources can be multiplied hundreds of
thousands of times.
Given this, the control plane protocol(s) implemented by OBPs to
provide the functionality discussed above should have the below
characteristics.
1. Minimize the amount of state needed to be stored on each OBP.
The OBP should only be required to cache state that it is
actively using, and be able to discard any cached state when it
is no longer required. For example, an OBP should only need to
maintain an inner-to-outer address mapping for destinations to
which it is actively sending traffic as opposed to maintaining
mappings for all possible destinations.
2. Fast acquisition of needed state. For example, when an End
Station emits a packet destined to an inner address that the OBP
does not have a mapping for, the OBP should be able to acquire
the needed mapping quickly.
3. Fast detection/update of stale cached state information. This
only applies if the cached state is actually being used. For
example, when a VM moves such that it is connected to a
different OBP, the inner to outer mapping for this VM's address
that is cached on other OBPs must be updated in a timely manner
(if they are actively in use). If the update is not timely, the
OBPs will forward data to the wrong OBP until it is updated.
Kreeger, et al. Expires August 2, 2012 [Page 10]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
4. Minimize processing overhead. This means that an OBP should
only be required to perform protocol processing directly related
to maintaining state for the End Stations it is actively
communicating with. This requirement is for the OBP
functionality only. The network node that contains the OBP may
be involved in other functionality for the underlying network
that maintains connectivity that the OBP is not actively using
(e.g., routing and multicast distribution protocols for the
underlying network).
5. Highly scalable. This means scaling to hundreds of thousands of
OBPs and several million VNs within a single administrative
domain. As the number of OBPs and/or VNs within a data center
grows, the protocol overhead at any one OBP should not increase
significantly.
6. Minimize the complexity of the implementation. This argues for
using the least number of protocols to achieve all the
functionality listed above. Ideally a single protocol should be
able to be used. The less complex the protocol is on the OBP,
the more likely interoperable implementations will be created in
a timely manner.
7. Extensible. The protocol should easily accommodate extension to
meet related future requirements. For example, access control
or QoS policies, or new address families for either inner or
outer addresses should be easy to add while maintaining
interoperability with OBPs running older versions.
8. Simple protocol configuration. A minimal amount of
configuration should be required for a new OBP to be
provisioned. Existing OBPs should not require any configuration
changes when a new OBP is provisioned. Ideally OBPs should be
able to auto configure themselves.
9. Do not rely on IP Multicast in the Underlying Network. Many
data centers do not have IP multicast routing enabled. If the
Underlying Network is an IP network, the protocol should allow
for, but not require the presence of IP multicast services
within the data center.
10. Flexible mapping sources. Inner to outer address mappings
should be able to be created by either the OBPs themselves or
other third party entities (e.g. data center management or
orchestration systems). The protocol should allow for mappings
created by an OBP to be automatically removed from all other
OBPs if it fails or is brought down unexpectedly.
Kreeger, et al. Expires August 2, 2012 [Page 11]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
11. Secure. See the Security Considerations section below.
5. Security Considerations
Editor's Note: This is an initial start on the security
considerations section; it will need to be expanded, and suggestions
for material to add are welcome.
The protocol(s) should protect the integrity of the mapping against
both off-path and on-path attacks. It should authenticate the
systems that are creating mappings, and rely on light weight security
mechanisms to minimize the impact on scalability and allow for simple
configuration.
Use of an overlay exposes virtual networks to attacks on the
underlying network beyond attacks on the control protocol that is the
subject of this draft. In addition to the directly applicable
security considerations for the networks involved, the use of an
overlay enables attacks on encapsulated virtual networks via the
underlying network. Examples of such attacks include traffic
injection into a virtual network via injection of encapsulated
traffic into the underlying network and modifying underlying network
traffic to forward traffic among virtual networks that should have no
connectivity. The control protocol should provide functionality to
help counter some of these attacks, e.g., distribution of OBP access
control lists for each virtual network to enable packets from non-
participating OBPs to be discarded, but the primary security measures
for the underlying network need to be applied to the underlying
network. For example, if the underlying network includes
connectivity across the public Internet, use of secure gateways
(e.g., based on IPsec [RFC 4301]) may be appropriate.
The inner to outer address mappings used for forwarding data towards
a remote OBP could also be used to filter incoming traffic to ensure
the inner address sourced packet came from the correct OBP source
address, allowing access control to discard traffic that does not
originate from the correct OBP. This destination filtering
functionality should be optional to use.
6. Acknowledgements
Thanks to the following people for reviewing and providing feedback:
Fabio Maino, Victor Moreno, Ajit Sanzgiri, Chris Wright.
Kreeger, et al. Expires August 2, 2012 [Page 12]
Internet-Draft Network Virtualization Overlay CP Reqs January 2012
Authors' Addresses
Lawrence Kreeger
Cisco
Email: kreeger@cisco.com
Dinesh Dutt
Cisco
Email: ddutt@cisco.com
Thomas Narten
IBM
Email: narten@us.ibm.com
David Black
EMC
Email: david.black@emc.com
Murari Sridharan
Microsoft
Email: muraris@microsoft.com
Kreeger, et al. Expires August 2, 2012 [Page 13]