| Internet-Draft | Available Session Recovery Protocol | January 2026 |
| Luo & Yan | Expires 27 July 2026 | [Page] |
This document describes an experimental protocol named the Available Session Recovery Protocol (ASRP). The protocol is designed to optimize high-availability network cluster architectures, providing a superior high-availability solution for clusters offering stateful network services such as load balancing and Network Address Translation (NAT [RFC4787] [RFC5508]). ASRP defines the procedures for session backup and recovery, as well as the message formats used during these interactions, enabling efficient and streamlined session state management.¶
In contrast to traditional high-availability techniques that back up session state within the cluster itself, the core innovation of ASRP lies in its distributed backup of state information to the client or server side. This approach offers multiple advantages: it significantly enhances the cluster's elastic scaling capabilities; supports rapid recovery from single-point or even multi-point failures; reduces resource redundancy by eliminating centralized backup nodes; and substantially simplifies the implementation complexity of the cluster.¶
The ASRP protocol provides exceptional elastic scalability for network clusters, facilitating the implementation and deployment of large-scale elastic network clusters.¶
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 https://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 27 July 2026.¶
Copyright (c) 2026 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 (https://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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Traditional high-availability network clusters based on a master-backup architecture rely on session state synchronization between the master and backup nodes. While functionally complete, this architecture faces challenges in the cloud era, such as insufficient flexibility for elastic scaling, resource redundancy, and high implementation complexity. To address these challenges, the industry has proposed the Elastic Stateful Cluster.¶
An Elastic Stateful Cluster is a high-availability network service cluster composed of multiple cooperative nodes. The number of nodes within the cluster can be elastically scaled, enabling it to provide stateful network services such as load balancing (SLB) and Network Address Translation (NAT). To achieve elastic scaling, traditional Elastic Stateful Clusters adopt a Fast/Slow Path design philosophy, separating session management from packet forwarding. This allows the fast path node layer to achieve good elastic scaling capabilities.¶
+--------------------------+
| +----------------------+ |
| | | |
| | Slow Path Nodes | |
| | | |
| +----------------------+ |
| ^ | |
| | | |
| +-------|--------|-----+ |
| | | ... | | |
+----------+ | | | ... V | | +----------+
| | | | +----------------+ | | | |
| Client | <--------> | Fast Path Node | <--------> | Server |
| | | | +----------------+ | | | |
+----------+ | | ... | | +----------+
| | ... | |
| +----------------------+ |
+--------------------------+
The slow path nodes are responsible for session creation and synchronization, while the fast path nodes are responsible for rapid packet forwarding. When a fast path node fails, external traffic can automatically switch to other healthy nodes, ensuring continuous service availability. The drawback of this Elastic Stateful Cluster architecture is the weak elastic scaling capability of the slow path nodes. Implementing session synchronization among slow path nodes is complex. A typical implementation reference is the AWS Hyperplane NFV platform.¶
+----------------------+
| ... |
+----------+ | ... | +----------+
| | | +----------------+ | | |
| Client | <--------> | ASRP Node | <--------> | Server |
| | | +----------------+ | | |
+----------+ | ... | +----------+
| ... |
+----------------------+
The Available Session Recovery Protocol (ASRP) proposes an innovative high-availability solution aimed at constructing a more concise, efficiently elastic, and highly available architecture for stateful services. Its core idea is to innovatively distribute backup session state information to the endpoints of a session (either the client or the server). The lifecycle of the backup state is strictly synchronized with the real session; it is created as the session is established and removed when the session terminates. By eliminating the need for independent keepalive and timeout mechanisms, this design ensures the timeliness and availability of the backup information. Based on this mechanism, network nodes within a cluster (such as load balancers or NAT devices) can rapidly reconstruct complete session states directly from the endpoints during node failures or cluster scaling events, thereby logically achieving a "stateless" nature for these network nodes.¶
To achieve the aforementioned objectives, ASRP defines corresponding session backup and recovery mechanisms. The protocol allows protocol messages to be transmitted together with a forwarded packet, thereby avoiding the overhead of additional control packets for state synchronization in the vast majority of cases.¶
In summary, by focusing on endpoint backup and rapid recovery of session state, the ASRP protocol effectively ensures session consistency and service continuity for services running on network nodes within a cluster. In an Elastic Stateful Cluster built using ASRP, network nodes possess atomic and mutually independent properties. There is no need for communication between nodes, nor is session synchronization required within the cluster. This fundamental design significantly enhances the cluster's elastic scaling capability, supports rapid recovery from single-point or even multi-point failures, and reduces resource redundancy and system implementation complexity by eliminating centralized backup nodes. Consequently, the ASRP protocol is particularly suitable for network environments that require frequent elastic scaling and pursue extremely high resource utilization and service stability, providing a robust solution for the deployment and operation of large-scale, highly elastic network clusters.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
For the ASRP protocol to function correctly, two prerequisites must be met. First, all network nodes within the cluster must run service software supporting the ASRP protocol. Second, the server or client responsible for backing up sessions must deploy a kernel module or an eBPF module that supports ASRP. Depending on whether this module is deployed on the server or the client, the protocol operates in one of two corresponding modes: Passive (PSV) Mode and Active (ACT) Mode.¶
In PSV mode, the network node is typically located within the same trusted network domain as the server (e.g., inside a data center). The network node backs up session state information to the server and recovers the session state from the server when necessary. This mode requires both the network node and the server to support ASRP. Its typical application scenarios include: traditional load balancer scenarios that provide services externally via a Virtual IP (VIP), or scenarios where an NFV load balancer network element provides cloud load balancing services.¶
In ACT mode, the network node is typically located within the same trusted network domain as the client (e.g., an enterprise intranet). The network node backs up session state information to the client and recovers the session state from the client when necessary. This mode requires both the network node and the client to support ASRP. Its typical application scenarios include: Source Network Address Translation (SNAT) scenarios, for example, where internal network devices access the internet via an SNAT gateway, or scenarios where cloud SNAT services are provided by an NFV SNAT network element.¶
Elastic
Stateful
Cluster
+------------------+
+----------+ | ... | +----------+
| | | +------------+ | | |
| Client | <----------> | node X | <----------> | Server |
| | | +------------+ | | |
+----------+ | ... | +----------+
+------------------+
Symmetric routing refers to a path type in which the bidirectional traffic of the same session between a client and a server is always routed to the same node within the cluster.¶
Typical examples that demonstrate symmetric routing include,¶
Active-Standby High Availability Architecture¶
All traffic for a specific session is always processed and maintained by the master node (e.g., NAT mapping tables, firewall session state). The backup node remains in a standby state, taking over only when the master node fails. This architecture inherently ensures symmetric routing of traffic to the master node.¶
Stateful Load Balancing Cluster with a "Same-Source-Same-Destination" Mechanism¶
In this modern, scaled architecture, network devices (such as OVS or routers) use a "same source, same destination" policy to ensure all packets belonging to the same connection are directed to the same load-balancing node, thereby maintaining symmetric routing in a distributed environment.¶
Elastic
Stateful
Cluster
+------------------+
| ... |
+----------+ | +------------+ | +----------+
| | -----------> | node X | -----------> | |
| | | +------------+ | | |
| Client | | ... | | Server |
| | | +------------+ | | |
| | <----------- | node Y | <----------- | |
+----------+ | +------------+ | +----------+
| ... |
+------------------+
Asymmetric routing refers to the scenario where bidirectional traffic of the same session may be routed to different nodes within a cluster. Specifically, a request from the client to the server might be processed by node X, while the corresponding response from the server to the client might be processed by node Y. This path inconsistency requires two or more nodes to collaboratively maintain the state information for the same session, posing a new challenge to the failure recovery mechanisms of stateful service clusters.¶
In cloud networking environments, asymmetric routing is an extremely common, even default, phenomenon. Taking an NFV network element cluster requiring elastic scaling as an example, one of the core architectural design goals is to allow nodes to scale out independently and flexibly. In such distributed architectures, unless specific traffic steering policies like "same source, same destination" are deployed, the underlying network devices (such as switches or load balancers) typically distribute traffic naturally and evenly across multiple available nodes based on mechanisms like ECMP (Equal-Cost Multi-Path [RFC2991], [RFC2992]), thereby commonly resulting in asymmetric routing.¶
ASRP achieves distributed backup and recovery of session state information by exchanging specific protocol messages among the client, server, and network nodes (such as load balancers or NAT devices). In a load-balancing scenario, session state is distributed and backed up to individual servers; in a Source Network Address Translation (SNAT) scenario, session state is distributed and backed up to individual clients.¶
ASRP defines four protocol messages: New Session message (NS), Hello Session message (HS), Query Session message (QS), and Recover Session message (RS). ASRP protocol messages are encapsulated within UDP ([RFC0768], [RFC3828]) datagrams for transmission. A specific destination port, referred to as ASRP-PORT (currently configurable for experimentation, e.g., 51200), is used to identify that the UDP payload contains an ASRP message.¶
A UDP datagram that carries an ASRP message is termed an ASRP packet (NS/HS/QS/RS packet). An ASRP packet can simultaneously carry both the ASRP message and the forwarded packet. If it carries only the ASRP message, it is referred to as a pure ASRP packet (pure NS/HS/QS/RS packet). If transmission can occur without causing IP fragmentation, the ASRP packet and the forwarded packet may be transmitted together; otherwise, to avoid IP fragmentation, the ASRP packet may be transmitted separately and given priority.¶
Generated by the network node, it is used to send session state information to a designated client (in ACT mode) or server (in PSV mode) for backup when creating a new session.¶
Generated by the client, it is used in ACT mode to announce to the network node its capability to support the ASRP protocol and to trigger the network node to return an NS message to complete session backup.¶
Generated by the network node, it is used to query the client or server for backup session state information when a received packet cannot match any local session and a session cannot be directly created. For TCP SYN packets, if no local session matches, a session can be created directly without querying the state.¶
Generated by the client or server holding the backup as a response to a QS message, it contains the state information required to recover the session. The network node parses the RS message and reconstructs the local session, thereby achieving failure recovery.¶
This section elaborates on, through a series of typical scenarios, how the ASRP protocol achieves session backup and recovery via message interaction in the event of network node failures under different operational modes. Each scenario details the involved protocol message flows and the processing steps of each entity.¶
Elastic
Stateful
Cluster
+----------+ +-------------+ +----------+
| | --1:PKT--> | | -----2:NS-----> | |
| | | | | |
| client | | Nodes | | server |
| | | | | |
| | <--4:PKT-- | | <----3:RS------ | |
+----------+ +-------------+ +----------+
a. recv PKT a. recv NS
b. new SESS b. new/get SESS
c. FWD NS c. send RS
d. recv RS
e. new/update SESS
f. FWD PKT
This scenario describes the direct session creation flow in PSV mode. The most common example is the SYN packet during TCP connection establishment, which represents the client initiating a new connection.¶
The processing flow is as follows:¶
Session Creation: Upon receiving a packet from a client (e.g., a TCP SYN), if no local session is found, the network node directly creates a new session. Subsequently, the network node sends an NS packet to the selected server. If the NS packet and the forwarded packet are transmitted separately, the NS packet is sent first.¶
Server Response: Upon receiving the NS message, the server backs up the session state information contained in the NS message locally and associates it with its local session. In the case of asymmetric routing, when the server sends its first response packet, it generates an RS packet and sends it to the network node.¶
Session Recovery: In the case of asymmetric routing, the network node, upon receiving the RS message, recovers the local session and subsequently forwards packets according to that session.¶
In the scenario above, provided that no IP fragmentation occurs, the NS/RS packets and the forwarded packets SHOULD be transmitted together to improve transmission efficiency. For example, for a TCP session, NS/RS messages are generally transmitted together with the SYN/SYN-ACK ([RFC9293]) packets. The backed-up session state information is released when the local session closes, requiring no additional close messages¶
Elastic
Stateful
Cluster
+----------+ +-------------+ +----------+
| | | | <----1:PKT---- | |
| | | | | |
| client | <--4:PKT--- | Nodes | -----2:QS----> | server |
| | | | | |
| | | | <----3:RS----- | |
+----------+ +-------------+ +----------+
a. recv PKT a. send PKT
b. no SESS b. recv QS
c. reply QS c. get SESS
d. recv RS d. reply RS
e. new SESS
f. FWD PKT
This scenario describes the session recovery flow triggered by a server packet when the network node has lost the session in PSV mode.¶
The processing flow is as follows:¶
Session Query: Upon receiving a packet from the server, the network node searches its local session table. If no corresponding session is found, the network node generates a QS packet and sends it back to the server.¶
Server-Assisted Reply: After receiving the QS packet, the server, based on the content of the QS message, looks up the locally stored backup session state information and then generates an RS packet, sending it back to the network node.¶
Session Recovery: After receiving the RS packet, the network node creates a new local session and subsequently forwards packets according to that session.¶
In the scenario above, provided that no IP fragmentation would occur, the forwarded packet may either be buffered or transmitted together with the QS packet; otherwise, the network node should buffer the forwarded packet. Once the session is recovered, any buffered packet must be processed immediately and forwarded in accordance with the session.¶
Elastic
Stateful
Cluster
+----------+ +-----------+ +------------+
| | | | ----2:QS----> | ... |
| | ---1:PKT--> | | <---3:RS----- | +--------+ |
| | | | ... | | server | |
| | | | ... | +--------+ |
| client | | Nodes | ----4:PKT---> | ... |
| | | | | +--------+ |
| | | | ----4:NS----> | | server | |
| | <--7:PKT--- | | <---5:RS----- | +--------+ |
| | | | <---6:RS----- | ... |
+----------+ +-----------+ +------------+
a. recv PKT a. recv QS
b. no SESS b. reply RS
c. send QS c. recv NS
d. recv RS d. new SESS
e. new/recover SESS e. reply RS
f. send NS f. send RS
g. recv RS
h. recover SESS
j. FWD PKT
This scenario describes the situation in PSV mode where, upon receiving a packet from a client, the network node cannot match it to a local session and cannot directly create a new session either. The network node must first determine whether this packet belongs to an existing session to decide how to handle it. The network node uses the ASRP protocol to query servers that may hold the backup session state information. ASRP relies on the cluster employing a deterministic server selection algorithm (such as a consistent hashing algorithm or a consistent hashing algorithm with history) to identify the target servers for querying.¶
A consistent hashing algorithm with history maintains a list of servers that have been used historically within a hash bucket. This list also serves as the target candidate server list for the network node's queries. ASRP recommends setting a maximum query count to avoid performance issues. Simultaneously, ASRP suggests setting a timeout for historical servers in the hash bucket to reduce the length of the server list by deleting timed-out historical records.¶
The processing flow is as follows:¶
Query Local Session: Upon receiving a forwarded packet from a client, the network node searches its local session table. If no local session is found, it calculates candidate servers (which may be multiple) using a deterministic server selection algorithm.¶
Query Backup Session: The network node sends QS packets to each candidate server to query for the backup session. The servers return the query results via RS packets.¶
Process Query Results: If a session is found, the network node recovers the local session based on the RS packet and then forwards the forwarded packet. If no session is found, it proceeds to the new session creation flow by sending an NS packet to the server selected by the algorithm.¶
Server Creates New Session: After receiving the NS packet, the server backs up the session state information locally. In an asymmetric routing environment, it must immediately reply with a pure RS packet as an acknowledgment.¶
Session Recovery: When the server sends its first service packet to the client, it generates an RS packet and sends it to the network node. Upon receiving the RS packet, the network node first recovers the local session based on the message and then forwards packets according to the session.¶
In the scenario above, provided that no IP fragmentation would occur, the forwarded packet may either be buffered or transmitted together with the QS packet; otherwise, the network node SHOULD buffer the forwarded packet. Once the session is created or recovered, any buffered packet must be processed immediately and forwarded in accordance with the session.¶
Elastic
Stateful
Cluster
+----------+ +-------------+ +----------+
| | -----1:HS----> | | | |
| | <----2:NS----- | | ---3:PKT--> | |
| client | | Nodes | | server |
| | <----5:QS----- | | <--4:PKT--- | |
| | -----6:RS----> | | | |
+----------+ +-------------+ +----------+
a. send HS a. recv HS
b. recv NS b. new session
c. store NS c. reply NS
d. recv QS d. recv PKT
e. reply RS e. no SESS
f. send QS
g. recv RS
h. FWD PKT
This scenario describes the session creation process by the network node and the server-packet-triggered session recovery flow in ACT mode. During the session recovery phase, the network node must be able to deterministically locate the client that holds the backup for that session. The use of a static, configurable mapping strategy is recommended. If such a mapping cannot be established, ASRP cannot function in this scenario. For SNAT services, ports can typically be used to map clients, with different clients using different, configurable port ranges.¶
The processing flow is as follows:¶
Session Creation: When a client initiates the first packet, it generates an HS packet and sends it to the network node. Upon receiving the HS packet, the network node follows the normal procedure to create a new session, returns a pure NS packet to the client, and forwards the forwarded packet according to the session.¶
Processing Server Response Packets: If a matching session is found, the packet is forwarded according to that session. If no matching session is found, the network node uses the mapping relationship to locate the corresponding client and sends a QS packet to it.¶
Client-Assisted Recovery: After receiving the QS packet, the client queries its locally stored backup session state information and replies with an RS packet to the network node.¶
Session Recovery: After receiving the RS packet, the network node recovers the session state locally and subsequently forwards packets according to the session.¶
After sending an HS message, the client waits for an NS message. If an NS message is not received, a minimum time interval (suggested on the order of milliseconds) is set. Subsequent packets sent by the client will trigger new HS messages to remind the network node to return an NS message. Upon receiving an HS message, if the network node does not find a matching local session, it creates a session, generates an NS message, and sends it to the client. If the network node subsequently receives further HS messages that do match a local session, it will also immediately send an NS message to the client.¶
In the scenario above, provided that no IP fragmentation would occur, the forwarded packet may either be buffered or transmitted together with the QS packet; otherwise, the network node SHOULD buffer the forwarded packet. Once the session is recovered, any buffered packet must be processed immediately and forwarded in accordance with the session.¶
Elastic
Stateful
Cluster
+----------+ +-------------+ +----------+
| | ---1:PKT---> | | | |
| | | | | |
| client | <---2:QS---- | Nodes | ----4:PKT---> | server |
| | | | | |
| | ----3:RS---> | | | |
+----------+ +-------------+ +----------+
a. send PKT a. recv PKT
b. recv QS b. no SESS
c. got SESS c. reply QS
d. replay RS d. recv RS
e. new SESS
f. FWD PKT
This scenario describes the client-packet-triggered session recovery flow in ACT mode.¶
The processing flow is as follows:¶
Session Query: Upon receiving a packet from a client, if the network node finds no local session and the packet does not contain an HS message, it sends a QS packet to the client.¶
Client-Assisted Recovery: After receiving the QS packet, the client queries its locally stored backup session state information and replies with an RS packet to the network node.¶
Session Recovery: After receiving the RS packet, the network node recovers the session locally and subsequently forwards the packet according to the session.¶
In the scenario above, provided that no IP fragmentation would occur, the forwarded packet may either be buffered or transmitted together with the QS packet; otherwise, the network node SHOULD buffer the forwarded packet. Once the session is recovered, any buffered packet must be processed immediately and forwarded in accordance with the session.¶
All ASRP protocol messages are encoded using the TLV (Type-Length-Value) structure. All numeric fields use network byte order (big-endian).¶
The fields that can be used in ASRP messages are as follows:¶
Sub and Type: 1 byte. Sub (high 4 bits) indicates the internal data type of the message; Type (low 4 bits) indicates the message type.¶
Length: 1 byte, indicating the total length of the entire ASRP message.¶
Flags: 1 byte. ASRP_F_ACT (0x1) is ACT mode flag; ASRP_F_MSG (0x2) is pure message flag.¶
Protocol: 1 byte, identifying the session protocol, such as TCP, UDP, etc.¶
Session-Tuple: Contains source address, destination address, source port, destination port. The IP address type is IPv4 or IPv6 ([RFC0791] [RFC8200]).¶
Session-Data: Variable-length field, carrying the private state information of the network node. The specific content is determined by the implementation and can generally be empty.¶
The NS message is used by the network node to back up session state information to the client or server. The NS message contains two Session-Tuples.¶
Type Assignments (Least significant nibble): NS: 0x0 Sub Assignments (Most significant nibble): ST44: 0x0, IPv4-Session-Tuple + IPv4-Session-Tuple ST66: 0x1, IPv6-Session-Tuple + IPv6-Session-Tuple ST46: 0x2, IPv4-Session-Tuple + IPv6-Session-Tuple ST64: 0x3, IPv6-Session-Tuple + IPv4-Session-Tuple¶
IPv4-Session-Tuple Format:¶
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Source IP (IPv4) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Destination IP (IPv4) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Source Port | Destination Port | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
IPv6-Session-Tuple Format: The structure of IPv6-Session-Tuple is the same as IPv4-Session-Tuple, with the main difference being the IP address length/type in the Session-Tuple field.¶
NS(Sub-ST44/ST66/ST46/ST64) Message Format:¶
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sub | Type | Length | Flags | Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ IPv4/IPv6/IPv4/IPv6-Session-Tuple ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ IPv4/IPv6/IPv6/IPv4-Session-Tuple ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Session-Data ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The NS message contains two Session-Tuples, representing the connection between the network node and the client, and the connection between the network node and the server, respectively.¶
The format of the NS/HS/QS/RS packet is as follows:¶
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ IP-Header + UDP Header (with destination port: ASRP-PORT) ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ NS/HS/QS/RS message ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Forwarded-PKT ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
If the ASRP_F_MSG flag is set in the Flags field of an NS/HS/QS/RS message, it indicates that this is a pure NS/HS/QS/RS packet (in which case the Forwarded-PKT section has a length of zero); otherwise, it indicates that the ASRP message packet is transmitted together with the forwarded packet.¶
The HS message is generated by the client to announce to the network node that it requires an NS message to back up session state information.¶
Type Assignments (Least significant nibble): HS: 0x1 Sub Assignments (Most significant nibble): NST: 0x0, No-Session-Tuple¶
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sub | Type | Length | Flags | reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The QS message is used by the network node to query backup session state information.¶
Type Assignments (Least significant nibble): 0x2: QS Sub Assignments (Most significant nibble): ST4: 0x0, IPv4-Session-Tuple ST6: 0x1, IPv6-Session-Tuple¶
QS(Sub-ST4/ST6) Message Format:¶
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sub | Type | Length | Flags | Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ IPv4/IPv6-Session-Tuple ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | ~ Session-Data ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The RS message is used to recover a network node's session.¶
Type Assignments (Least significant nibble): 0x3: RS Sub Assignments (Most significant nibble): ST44: 0x0, IPv4-Session-Tuple + IPv4-Session-Tuple ST66: 0x1, IPv6-Session-Tuple + IPv6-Session-Tuple ST46: 0x2, IPv4-Session-Tuple + IPv6-Session-Tuple ST64: 0x3, IPv6-Session-Tuple + IPv4-Session-Tuple ST4: 0x4, IPv4-Session-Tuple ST6: 0x5, IPv6-Session-Tuple¶
RS(Sub-ST44/ST66/ST46/ST64) Message Format: The structure of RS(ST44/ST66/ST46/ST64) messages is the same as NS(ST44/ST66/ST46/ST64).¶
RS(Sub-ST4/ST6) Message Format: The structure of RS(ST4/ST6) messages is the same as QS(ST4/ST6).¶
If the Sub field of an RS message is ST44, ST66, ST46, or ST64, it indicates the RS message carries session recovery information. If the Sub field is ST4 or ST6, this RS message is a response indicating a failed query for the corresponding QS message.¶
ASRP packets can be identified by their UDP destination port. Once an ASRP packet is identified, the ASRP message within the packet can be parsed and processed.¶
The NS message is generated by a network node when creating a new session and is used to back up the session to the client or server.¶
The source IP of the NS packet is set to the network node's local IP (which can be obtained from configuration), and the destination IP is set to the client's or server's IP (which can be obtained from the forwarded packet). The source port is randomly generated, and the destination port is set to ASRP-PORT. If it can be done without causing fragmentation, the forwarded IP packet can be placed after the NS message and transmitted together with it. If the NS message is transmitted alone, the ASRP_F_MSG flag must be set in the NS message's Flags field.¶
In PSV mode, if an NS message is lost, for TCP connections, the retransmission of the SYN packet will trigger the retransmission of the NS message. For other types of connections, subsequent packets will continue to generate NS messages until an RS message is received.¶
In ACT mode, if an NS message is lost, subsequent packets sent by the client will generate HS messages, prompting the network node to retransmit the NS message in response to these subsequent HS messages.¶
NS messages may be generated in both PSV and ACT modes. The handling procedures are described in Figure 5, Figure 7, and Figure 8.¶
The HS message is sent by the client during the initial connection establishment phase to announce to the network node that it requires an NS message to back up session state information.¶
The source IP, destination IP, and source port for the HS packet are copied from the sending packet; the destination port is set to ASRP-PORT. If it can be done without causing fragmentation, the forwarded IP packet can be placed after the HS message and transmitted together with it. If the HS message is transmitted alone, the ASRP_F_MSG flag must be set in the HS message's Flags field.¶
HS messages are only generated in ACT mode. The handling procedure is described in Figure 8.¶
The QS message is generated by the network node to query backup session state information.¶
The source IP of the QS packet is set to the network node's local IP (obtainable from configuration), and the destination IP is set to the client's or server's IP (obtainable from the forwarded packet as described in Figure 6 and Figure 9, or derived via algorithmic mapping to the client or server as described in Figure 7 and Figure 8). The source port is randomly generated, and the destination port is set to ASRP-PORT. If transmission can occur without causing fragmentation, the forwarded IP packet may be appended after the QS message and transmitted together. If the QS message is transmitted independently, the ASRP_F_MSG flag must be set in the Flags field of the QS message.¶
If a QS message is lost, subsequent packets will trigger the generation of new QS packets, continuing the attempt to recover the session.¶
QS messages may be generated in both PSV and ACT modes. The handling procedures are described in Figure 6, Figure 7, and Figure 9.¶
The RS message is generated by the client or server in response to an NS or QS message. It is processed by the network node to recover a session.¶
When an RS message responds to a QS message, the RS packet reuses the IP/UDP header of the QS packet, with the IP addresses and UDP ports swapped (source and destination are exchanged). When an RS message responds to an NS message, the RS packet is triggered by the currently sent packet. It reuses the source and destination IPs of the current packet, with a randomly generated source port and the destination port set to ASRP-PORT. If transmission can occur without causing fragmentation, the forwarded IP packet may be appended after the RS message and transmitted together. If the RS message is transmitted independently, the ASRP_F_MSG flag must be set in the Flags field of the RS message.¶
When a client or server receives a QS message, it searches locally for the backup session state information. If found, it generates an RS message (with Sub-ST44, ST66, ST46, or ST64) and sends it back to the network node. If the corresponding session state information is not found, it returns an RS message (with Sub-ST4 or ST6) to the network node.¶
If an RS message is lost, subsequent NS or QS messages will continue the attempt to recover the session, thereby triggering retransmission of the RS message.¶
RS messages may be generated in both PSV and ACT modes. The handling procedures are described in Figure 5, Figure 6, Figure 7, Figure 8, and Figure 9.¶
The security design of the ASRP protocol is based on its typical deployment model.¶
Deployment Boundaries and Access Control: ASRP recommends deploying network nodes and the clients or servers that back up sessions within the same trusted internal network domain. In this model, all ASRP protocol packets communicate within an internal address space. By implementing appropriate network segmentation (e.g., using firewall policies or security groups) and strictly checking the source addresses of packets, forged ASRP packets originating from untrusted external networks can be effectively prevented from reaching the target nodes.¶
Session Legitimacy Verification: When processing ASRP packets that may establish new sessions (e.g., HS or RS packets), network nodes MUST perform basic validation according to the specific policies of the upper-layer application or service. For instance, in a load-balancing scenario, a node SHOULD verify whether the session points to a known and healthy server. In a NAT scenario, it SHOULD verify whether the address translation complies with predefined rules. This prevents the establishment of illegal sessions at the application layer.¶
Internal Threat Assessment: Even if an attacker is located within the trusted network and can forge ASRP packets, the scope of impact is inherently limited. The attacker can only forge sessions where they themselves are the endpoint (e.g., masquerading as a client to request recovery of a non-existent connection). Such forged sessions are indistinguishable in form from sessions established through normal access. They do not directly jeopardize the security of other users or nodes, nor can they elevate the attacker's privileges or grant access to unauthorized resources.¶
When a network node loses a session, it may generate a large volume of QS packets. If maliciously exploited or due to a malfunction, this could lead to a flood attack [RFC4987]. To mitigate such risks, implementers should consider the following protective measures:¶
Rate Limiting and Traffic Shaping: Each network node MUST implement monitoring and limiting of the rate at which QS packets are sent. A reasonable threshold (e.g., the number of QS packets allowed per second) SHOULD be set. When the rate exceeds this threshold, the node SHOULD adopt a packet drop policy, for example, discarding newly arriving service packets that trigger queries. The parameters for rate limiting SHOULD be configurable to adapt to deployment environments of different scales.¶
This document defines an application-layer protocol (ASRP). The protocol message types and internal identifiers are defined by this specification itself and constitute internal implementation details of the protocol. Therefore, there is no need to request registration of a separate protocol number or code point from IANA. However, for the implementation of this protocol, a UDP destination port requires allocation:¶
ASRP messages (NS, HS, QS, RS) are encapsulated within UDP datagrams for transmission. A fixed UDP destination port number is required so that the receiving end can identify and process such encapsulated packets.¶
Service Name: asrp¶
Port Number: 51200 (proposed value for current experimentation)¶
Transport Protocol: udp¶
Description: Used for receiving UDP-encapsulated ASRP protocol messages.¶
For experimental implementations and interoperability testing prior to IANA assignment, UDP port 51200 MAY be used as a temporary default. This port falls within the dynamic/private port range (49152-65535) reserved for local or temporary use and documentation examples [RFC6335].¶
IANA is requested to assign a permanent port number in the "User Ports" range (1024-49151) for the "asrp" service in the "Service Name and Transport Protocol Port Number Registry", with a reference to this document.¶
The authors would like to thank all individuals who have provided valuable feedback and contributions during the development of this document.¶