| Network Working Group | J. Gross |
| Internet-Draft | T. Sridhar |
| Intended status: Informational | VMware |
| Expires: August 18, 2014 | P. Garg |
| Microsoft | |
| C. Wright | |
| Red Hat | |
| I. Ganga | |
| Intel | |
| February 14, 2014 |
Genève: Generic Network Virtualization Encapsulation
draft-gross-geneve-00
Network virtualization involves the cooperation of devices with a wide variety of capabilities such as software and hardware tunnel endpoints, transit fabrics, and centralized control clusters. As a result of their role in tying together different elements in the system, the requirements on tunnels are influenced by all of these components. Flexibility is therefore the most important aspect of a tunnel protocol if it is keep pace with the evolution of the system. This draft describes Genève, a protocol designed to recognize and accommodate these changing capabilities and needs.
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 18, 2014.
Copyright (c) 2014 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 (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.
Networking has long featured a variety of tunneling, tagging, and other encapsulation mechanisms. However, the advent of network virtualization has caused a surge of renewed interest and a corresponding increase in the introduction of new protocols. The large number of protocols in this space, ranging all the way from VLANs [IEEE.802.1Q-2011] and MPLS [RFC3031] through the more recent VXLAN [I-D.mahalingam-dutt-dcops-vxlan], NVGRE [I-D.sridharan-virtualization-nvgre], and STT [I-D.davie-stt], often leads to questions about the need for new encapsulation formats and what it is about network virtualization in particular that leads to their proliferation.
While many encapsulation protocols seek to simply partition the underlay network or bridge between two domains, network virtualization views the transit network as providing connectivity between multiple components of an integrated system. In many ways this system is similar to a chassis switch with the IP underlay network playing the role of the backplane and tunnel endpoints on the edge as line cards. When viewed in this light, the requirements placed on the tunnel protocol are significantly different in terms of the quantity of metadata necessary and the role of transit nodes.
Current work such as [VL2] and the NVO3 working group [I-D.ietf-nvo3-dataplane-requirements] have described some of the properties that the data plane must have to support network virtualization. However, one additional defining requirement is the need to carry system state along with the packet data. The use of some metadata is certainly not a foreign concept - nearly all protocols used for virtualization have at least 24 bits of identifier space as a way to partition between tenants. This is often described as overcoming the limits of 12-bit VLANs, and when seen in that context, or any context where it is a true tenant identifier, 16 million possible entries is a large number. However, the reality is that the metadata is not exclusively used to identify tenants and encoding other information quickly starts to crowd the space. In fact, when compared to the tags used to exchange metadata between line cards on a chassis switch, 24-bit identifiers start to look quite small. There are nearly endless uses for this metadata, ranging from storing input ports for simple security policies to service based context for interposing advanced middleboxes.
Existing tunnel protocols have each attempted to solve different aspects of these new requirements, only to be quickly rendered out of date by changing control plane implementations and advancements. Furthermore, software and hardware components and controllers all have different advantages and rates of evolution - a fact that should be viewed as a benefit, not a liability or limitation. This draft describes Genève, a protocol which seeks to avoid these problems by providing a framework for tunneling rather than being prescriptive about the entire system.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].
In this document, these words will appear with that interpretation only when in ALL CAPS. Lower case uses of these words are not to be interpreted as carrying RFC-2119 significance.
The following terms are used in this document:
Checksum offload. An optimization implemented by many NICs which enables computation and verification of upper layer protocol checksums in hardware on transmit and receive, respectively. This typically includes IP and TCP/UDP checksums which would otherwise be computed by the protocol stack in software.
Clos network. A technique for composing network fabrics larger than a single switch while maintaining non-blocking bandwidth across connection points. ECMP is used to divide traffic across the multiple links and switches that constitute the fabric. Sometimes termed "leaf and spine" or "fat tree" topologies.
ECMP. Equal Cost Multipath. A routing mechanism for selecting from among multiple best next hop paths by hashing packet headers in order to better utilize network bandwidth while avoiding reordering a single stream.
Genève. Generic Network Virtualization Encapsulation. The tunnel protocol described in this draft.
LRO. Large Receive Offload. The receive-side equivalent function of LSO, in which multiple protocol segments (primarily TCP) are coalesced into larger data units.
NIC. Network Interface Card. A NIC could be part of a tunnel endpoint or transit device and can either process Genève packets or aid in the processing of Genève packets.
OAM. Operations, Administration, and Management. A suite of tools used to monitor and troubleshoot network problems.
Transit device. A forwarding element along the path of the tunnel. A transit device MAY be capable of understanding the Genève frame format but does not originate or terminate Genève packets.
LSO. Large Segmentation Offload. A function provided by many commercial NICs that allows data units larger than the MTU to be passed to the NIC to improve performance, the NIC being responsible for creating smaller segments with correct protocol headers. When referring specifically to TCP/IP, this feature is often known as TSO (TCP Segmentation Offload).
Tunnel endpoint. A component encapsulating Ethernet frames in Genève headers and vice versa. As the ultimate consumer of any tunnel metadata, endpoints have the highest level of requirements for parsing and interpreting tunnel headers. Tunnel endpoints may consist of either software or hardware implementations or a combination of the two.
VM. Virtual Machine.
Genève is designed to support network virtualization use cases, where tunnels are typically established to act as a backplane between the virtual switches residing in hypervisors, physical switches, or middleboxes or other appliances. An arbitrary IP network can be used as an underlay although Clos networks composed using ECMP links are a common choice to provide consistent bisectional bandwidth across all connection points. Figure 1 shows an example of a hypervisor, top of rack switch for connectivity to physical servers, and a WAN uplink connected using Genève tunnels over a simplified Clos network. These tunnels are used to encapsulate and forward frames from the attached components such as VMs or physical links.
+---------------------+ +-------+ +------+
| +--+ +-------+---+ | |Transit|--|Top of|==Physical
| |VM|--| | | | +------+ /|Router | | Rack |==Servers
| +--+ |Virtual|NIC|---|Top of|/ +-------+\/+------+
| +--+ |Switch | | | | Rack |\ +-------+/\+------+
| |VM|--| | | | +------+ \|Transit| |Uplink| WAN
| +--+ +-------+---+ | |Router |--| |=========>
+---------------------+ +-------+ +------+
Hypervisor
()===================================()
Switch-Switch Geneve Tunnels
Figure 1: Sample Genève Deployment
In order to support the needs of network virtualization, the tunnel protocol should be able to take advantage of the differing (and evolving) capabilities of each type of device in both the underlay and overlay networks. This results in the following requirements being placed on the data plane tunneling protocol:
These requirements are described further in the following subsections.
Although some protocols for network virtualization have included a control plane as part of the tunnel format specification (most notably, the original VXLAN spec prescribed a multicast learning-based control plane), these specifications have largely been treated as describing only the data format. The VXLAN frame format has actually seen a wide variety of control planes built on top of it.
There is a clear advantage in settling on a data format: most of the protocols are only superficially different and there is little advantage in duplicating effort. However, the same cannot be said of control planes, which are diverse in very fundamental ways. The case for standardization is also less clear given the wide variety in requirements, goals, and deployment scenarios.
As a result of this reality, Genève aims to be a pure tunnel format specification that is capable of fulfilling the needs of many control planes by explicitly not selecting any one of them. This simultaneously promotes a shared data format and increases the chances that it will not be obsoleted by future control plane enhancements.
Achieving this level of flexibility effectively requires an options infrastructure to allow new metadata types to be defined, deployed, and either finalized or retired. Options also allow for differentiation of products by encouraging independent development in each vendor's core specialty, leading to an overall faster pace of advancement. By far the most common mechanism for implementing options is Type-Length-Value (TLV) format.
It should be noted that while options can be used to support non-wirespeed control frames, they are equally important on data frames as well to segregate and direct forwarding (for instance, the examples given before of input port based security policies and service interposition both require tags to be placed on data packets). Therefore, while it would be desirable to limit the extensibility to only control frames for the purposes of simplifying the datapath, that would not satisfy the design requirements.
There is often a conflict between software flexibility and hardware performance that is difficult to resolve. For a given set of functionality, it is obviously desirable to maximize performance. However, that does not mean new features that cannot be run at that speed today should be disallowed. Therefore, for a protocol to be efficiently implementable means that a set of common capabilities can be reasonably handled across platforms along with a graceful mechanism to handle more advanced features in the appropriate situations.
The use of a variable length header and options in a protocol often raises questions about whether it is truly efficiently implementable in hardware. To answer this question in the context of Genève, it is important to first divide "hardware" into two categories: tunnel endpoints and transit devices.
Endpoints must be able to parse the variable header, including any options, and take action. Since these devices are actively participating in the protocol, they are the most affected by Genève. However, as endpoints are the ultimate consumers of the data, transmitters can tailor their output to the capabilities of the recipient. As new functionality becomes sufficiently well defined to add to endpoints, supporting options can be designed using ordering restrictions and other techniques to ease parsing.
Transit devices MAY be able to interpret the options and participate in Genève packet processing. However, as non-terminating devices, they do not originate or terminate the Genève packet. The participation of transit devices in Genève packet processing is OPTIONAL.
Further, either tunnel endpoints or transit devices MAY use offload capabilities of NICs such as checksum offload to improve the performance of Genève packet processing. The presence of a Genève variable length header SHOULD NOT prevent the tunnel endpoints and transit devices from using such offload capabilities.
IP has clearly cemented its place as the dominant transport mechanism and many techniques have evolved over time to make it robust, efficient, and inexpensive. As a result, it is natural to use IP fabrics as a transit network for Genève. Fortunately, the use of IP encapsulation and addressing is enough to achieve the primary goal of delivering packets to the correct point in the network through standard switching and routing.
In addition, nearly all underlay fabrics are designed to exploit parallelism in traffic to spread load across multiple links without introducing reordering in individual flows. These equal cost multipathing (ECMP) techniques typically involve parsing and hashing the addresses and port numbers from the packet to select an outgoing link. However, the use of tunnels often results in poor ECMP performance without additional knowledge of the protocol as the encapsulated traffic is hidden from the fabric by design and only endpoint addresses are available for hashing.
Since it is desirable for Genève to perform well on these existing fabrics, it is necessary for entropy from encapsulated packets to be exposed in the tunnel header. The most common technique for this is to use the UDP source port, which is discussed further in Section 3.3.
The Genève frame format consists of a compact tunnel header encapsulated in UDP over either IPv4 or IPv6. A small fixed tunnel header provides control information plus a base level of functionality and interoperability with a focus on simplicity. This header is then followed by a set of variable options to allow for future innovation. Finally, the payload consists of a protocol data unit of the indicated type, such as an Ethernet frame. The following subsections provide examples of Genève frames transported (for example) over Ethernet along with an Ethernet payload.
0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 Outer Ethernet Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Destination MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Destination MAC Address | Outer Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Optional Ethertype=C-Tag 802.1Q| Outer VLAN Tag Information | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Ethertype=0x0800 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Outer IPv4 Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Version| IHL |Type of Service| Total Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Identification |Flags| Fragment Offset | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Time to Live |Protocol=17 UDP| Header Checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Source IPv4 Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Destination IPv4 Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Outer UDP Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Source Port = xxxx | Dest Port = Geneve Port | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | UDP Length | UDP Checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Geneve Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Ver| Opt Len |O|C| Rsvd. | Protocol Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Virtual Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Variable Length Options | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Inner Ethernet Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Inner Destination MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Inner Destination MAC Address | Inner Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Inner Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Optional Ethertype=C-Tag 802.1Q| Inner VLAN Tag Information | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Payload: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Ethertype of Original Payload | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | Original Ethernet Payload | | | | (Note that the original Ethernet Frame's FCS is not included) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Frame Check Sequence: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | New FCS (Frame Check Sequence) for Outer Ethernet Frame | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 Outer Ethernet Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Destination MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Destination MAC Address | Outer Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Outer Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Optional Ethertype=C-Tag 802.1Q| Outer VLAN Tag Information | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Ethertype=0x86DD | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Outer IPv6 Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Version| Traffic Class | Flow Label | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Payload Length | NxtHdr=17 UDP | Hop Limit | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + + | | + Outer Source IPv6 Address + | | + + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | + + | | + Outer Destination IPv6 Address + | | + + | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Outer UDP Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Source Port = xxxx | Dest Port = Geneve Port | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | UDP Length | UDP Checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Geneve Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Ver| Opt Len |O|C| Rsvd. | Protocol Type | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Virtual Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Variable Length Options | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Inner Ethernet Header: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Inner Destination MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Inner Destination MAC Address | Inner Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Inner Source MAC Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Optional Ethertype=C-Tag 802.1Q| Inner VLAN Tag Information | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Payload: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Ethertype of Original Payload | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | Original Ethernet Payload | | | | (Note that the original Ethernet Frame's FCS is not included) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Frame Check Sequence: +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | New FCS (Frame Check Sequence) for Outer Ethernet Frame | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The use of an encapsulating UDP [RFC0768] header follows the connectionless semantics of Ethernet and IP in addition to providing entropy to routers performing ECMP. The header fields are therefore interpreted as follows:
Transit devices MUST maintain consistent forwarding behavior irrespective of the value of 'Opt Len', including ECMP link selection. These devices SHOULD be able to forward packets containing options without resorting to a slow path.
0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Option Class | Type |R|R|R| Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Variable Option Data | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Genève Option
The base Genève header is followed by zero or more options in Type-Length-Value format. Each option consists of a four byte option header and a variable amount of option data interpreted according to the type.
Genève options are intended to be originated and processed by tunnel endpoints. Options MAY be processed by transit devices along the tunnel path as well. This document only details the handling of options by tunnel endpoints. A future version of this document will provide details of options processing by transit devices. Transit devices not processing Genève options SHOULD process Genève frame as any other UDP frame and maintain consistent forwarding behavior.
In tunnel endpoints, the generation and interpretation of options is determined by the control plane, which is out of the scope of this document. However, to ensure interoperability between heterogeneous devices two requirements are imposed on endpoint devices:
As an IP-based tunnel protocol, Genève shares many properties and techniques with existing protocols. The application of some of these are described in further detail, although in general most concepts applicable to the IP layer or to IP tunnels generally also function in the context of Genève.
In order to prevent fragmentation and maximize performance, the best practice when using Genève is to ensure that the MTU of the physical network is greater than or equal to the MTU of the encapsulated network plus tunnel headers. Manual or upper layer (such as TCP MSS clamping) configuration can be used to ensure that fragmentation never takes place, however, in some situations this may not be feasible.
It is RECOMMENDED that Path MTU Discovery be used by setting the DF bit in the IP header when Genève packets are transmitted over IPv4 (this is the default with IPv6). The use of Path MTU Discovery on the transit network provides the encapsulating endpoint with soft-state about the link that it may use to prevent or minimize fragmentation depending on its role in the virtualized network.
If necessary, it is RECOMMENDED that fragmentation be performed preferentially on the encapsulated payload. This may be possible if the encapsulating endpoint is also acting as an L3 node in the virtualized network, in which case the endpoint might use the derived transit MTU and the tunnel header length to either implement Path MTU Discovery or fragment the inner packet to the correct size.
In many cases it may not be possible or desirable for the tunnel endpoint to interact with the payload, such as when implementing a completely transparent L2 bridge. In these situations, fragmentation of the transit IP header MAY be performed to ensure connectivity. If a packet is fragmented endpoints SHOULD use the path MTU of the transit link to ensure a size is chosen such that fragmentation is only required once between endpoints. Note that some implementations may not be capable of supporting fragmentation or other less common features of the IP header, such as options and extension headers.
When encapsulating IP (including over Ethernet) frames in Genève, there are several options for propagating DSCP and ECN bits from the inner header to the tunnel on transmission and the reverse on reception.
[RFC2983] lists considerations for mapping DSCP between inner and outer IP headers. Network virtualization is typically more closely aligned with the Pipe model described, where the DSCP value on the tunnel header is set based on a policy (which may be a fixed value, one based on the inner traffic class, or some other mechanism for grouping traffic). Aspects of the Uniform model (which treats the inner and outer DSCP value as a single field by copying on ingress and egress) may also apply, such as the ability to remark the inner header on tunnel egress based on transit marking. However, the Uniform model is not conceptually consistent with network virtualization, which seeks to provide strong isolation between encapsulated traffic and the physical network.
[RFC6040] describes the mechanism for exposing ECN capabilities on IP tunnels and propagating congestion markers to the inner packets. This behavior SHOULD be followed for IP packets encapsulated in Genève.
Genève tunnels may either be point-to-point unicast between two endpoints or may utilize broadcast or multicast addressing. It is not required that inner and outer addressing match in this respect. For example, in physical networks that do not support multicast, encapsulated multicast traffic may be replicated into multiple unicast tunnels or forwarded by policy to a unicast location (possibly to be replicated there).
With physical networks that do support multicast it may be desirable to use this capability to take advantage of hardware replication for encapsulated packets. In this case, multicast addresses may be allocated in the physical network corresponding to tenants, encapsulated multicast groups, or some other factor. The allocation of these groups is a component of the control plane and therefore outside of the scope of this document. When physical multicast is in use, the 'C' bit in the Genève header may be used with groups of devices with heterogeneous capabilities as each device can interpret only the options that are significant to it if they are not critical.
Modern NICs currently provide a variety of offloads to enable the efficient processing of packets. The implementation of many of these offloads requires only that the encapsulated packet be easily parsed (for example, checksum offload). However, optimizations such as LSO and LRO involve some processing of the options themselves since they must be replicated/merged across multiple packets. In these situations, it is desirable to not require changes to the offload logic to handle the introduction of new options. To enable this, some constraints are placed on the definitions of options to allow for simple processing rules:
There is no requirement that a given implementation of Genève employ the offloads listed as examples above. However, as these offloads are currently widely deployed in commercially available NICs, the rules described here are intended to enable efficient handling of current and future options across a variety of devices.
Genève is capable of encapsulating a wide range of protocols and therefore a given implementation is likely to support only a small subset of the possibilities. However, as Ethernet is expected to be widely deployed, it is useful to describe the behavior of VLANs inside encapsulated Ethernet frames.
As with any protocol, support for inner VLAN headers is OPTIONAL. In many cases, the use of encapsulated VLANs may be disallowed due to security or implementation considerations. However, in other cases trunking of VLAN frames across a Genève tunnel can prove useful. As a result, the processing of inner VLAN tags upon ingress or egress from a tunnel endpoint is based upon the configuration of the endpoint and/or control plane and not explicitly defined as part of the data format.
Viewed exclusively from the data plane, Genève does not introduce any interoperability issues as it appears to most devices as UDP frames. However, as there are already a number of tunnel protocols deployed in network virtualization environments, there is a practical question of transition and coexistence.
Since Genève is a superset of the functionality of the three most common protocols used for network virtualization (VXLAN, NVGRE, and STT) it should be straightforward to port an existing control plane to run on top of it with minimal effort. With both the old and new frame formats supporting the same set of capabilities, there is no need for a hard transition - endpoints directly communicating with each other use any common protocol, which may be different even within a single overall system. As transit devices are primarily forwarding frames on the basis of the IP header, all protocols appear similar and these devices do not introduce additional interoperability concerns.
In order to assist with this transition, it is strongly suggested that implementations support simultaneous operation of both Genève and existing tunnel protocols as it is expected to be common for a single node to communicate with a mixture of other nodes. Eventually, older protocols may be phased out as they are no longer in use.
As UDP/IP packets, Genève does not have any inherent security mechanisms. As a result, an attacker with access to the underlay network transporting the IP frames has the ability to snoop or inject packets. Legitimate but malicious tunnel endpoints may also spoof identifiers in the tunnel header to gain access to networks owned by other tenants.
Within a particular security domain, such as a data center operated by a single provider, the most common and highest performing security mechanism is isolation of trusted components. Tunnel traffic can be carried over a separate VLAN and filtered at any untrusted boundaries. In addition, tunnel endpoints should only be operated in environments controlled by the service provider, such as the hypervisor itself rather than within a customer VM.
When crossing an untrusted link, such as the public Internet, IPsec [RFC4301] may be used to provide authentication and/or encryption of the IP packets. If the remote tunnel endpoint is not completely trusted, for example it resides on a customer premises, then it may also be necessary to sanitize any tunnel metadata to prevent tenant-hopping attacks.
A UDP destination port in the user range (1024-49151) will be requested from IANA.
In addition, IANA will be requested to create a "Genève Option Class" registry to allocate Option Classes. This shall be a registry of 16-bit hexadecimal values along with descriptive strings. The identifiers 0x0-0xFF are to be reserved for standardized options for allocation by an IETF working group document or RFC and 0xFFFF for experimental use. Otherwise, identifiers are to be assigned to any organization with an interest in creating Genève options on a first come first serve basis.
The authors wish to thank Martìn Casado, Bruce Davie and Dave Thaler for their input, feedback, and helpful suggestions.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [RFC0768] | Postel, J., "User Datagram Protocol", STD 6, RFC 768, August 1980. |