Signaling In-Network Computing operations (SINC) deployment considerations

Introduction "Signaling In-Network Computing operations" (SINC) is a mechanism to enable signaling in-network computing operations on data packets in specific scenarios like NetReduce, NetDistributedLock, NetSequencer, etc. This mechanism can effectively reduce the task completion time and improve the system efficiency. The SINC framework design can be found in the draft .

Terminology This document uses the terms as defined in , , , , and . This document assume that the reader is familiar with the Service Function Chaining architecture and Multi-Protocol Label Switching architecture.

SINC Deployment Considerations In order to deploy the SINC solution in the network, the packet from the host needs to be transmitted through a path containing SINC capable switches/routers (SW/R) for data computation. Different from the simplistic experimental network environment used for in network computing verification in research papers, the real network is much more complex, containing multiple SINC SW/Rs with different capabilities and multiple paths between sources and destinations, as shown in . The packet traveling between Host A and Host B may pass SINC Node A or B via different paths. It is essential to create a proper route with nodes to support SINC operation and facilitate the packet delivery. The SINC capable SW/Rs should periodically advertise their networking & computing capacities and capabilities, e.g. the operation it can perform, the current work load, the link capacity, etc. Based on those information, the control plane is responsible to create a proper route where the data in the packet will undertake the desired computation before arriving at the destination host. Such a path could be located at layer 2, 3 or 4 dependent of the network context and application environments. For instance, in a telecommunication network where the Multi-Protocol Label Switching (MPLS) is deployed, we could used the MPLS protocol to encapsulate the SINC header and deliver the packet to a SINC capable SW/R. In a Data Center Network, if the Generic Network Virtualization Encapsulation (GENEVE) is applied, we may use the GENEVE protocol for encapsulation. Other encapsulation protocols like General Routing Encapsulation (GRE) , Service Function Chaining (SFC) , and so on, could be potential candidates as well. The SINC header is usually copied/moved right after the new encapsulation header, which makes it easier to access the SINC header.

SINC In Deployment Topology

SINC over SFC Considerations In this section, SFC, which is a layer 3.5 protocol, is used as a running example on how to create a tunnel and encapsulate the SINC header, in order to implement the desired in-network computation. shows the architecture of a SFC-based SINC network. In the computing service chain, a host sends out packets containing data operations to be executed in the network. The data operation description should be carried in the packet itself by using a SINC-specific NSH encapsulation. Once the SINC packet enters into the SFC domain, the Service Function Forwarder (SFF) will forward packets to one or more connected service functions according to information carried in the SFC encapsulation. The Service Function (SF) is responsible for implementing data operations.

SINC over SFC Architecture. | SFF |-->| | SFF | |-->| SFF |-->| Proxy | +------------+ +-----+ | +-----+ | +-----+ +-----------+ | | | | +-----+ | | | SF | | | +-----+ | +-----------+ ]]>

SINC NSH encapsulation This section shows how the SINC header can be embedded in the Network Service Header (NSH) for SFC . The SINC NSH base header, as shown in , is basically another type of NSH Meta Data (MD) header. SINC NSH encapsulation uses the NSH MD fixed-length context headers to carry the data operation information as show in . Please refer to the NSH for a detailed SFC basic header description.

NSH Base Header, where 'MD Type' should contain a code-point assigned by IANA. Following the NSH basic header there is the Service Path Header, show in , as defined in .

NSH Service Path Header. The complete SINC NSH header, as shown in , stacks the NSH base header, NSH Service Path Header, and the SINC Header all together.

SINC NSH Header.

SINC over SFC Workflow For the sake of clarity, a simple example with one sender (Host A) and one receiver (Host B) is used to illustrate the workflow. Packet processing goes through the following steps:

Host A transmits a packet containing a SINC header together with data, which will be processed by SFs on SINC capable SW/Rs, to the SINC Ingress Proxy.
The SINC Ingress Proxy encapsulates the original packet with the SINC header into a SINC NSH header, as the transport protocol indicated by the SFC.
The SFF forwards the encapsulated packet to the SINC SW/R.
As shown in , when the packet reaches the SINC SW/R, the header of the packet will be removed. The SFF looks up the Service Path Identifier (SPI) table and Service Index (SI) table and sends the packet to the SF.
The SF performs the computing based on the defined operation in the SINC header after the verification of the Group ID and Data Source ID. The payload will be replaced with the result after computation. The Operation Done flag will be set to 1. The packet is then re-encapsulated with the NSH SINC header. The SI is reduce by 1 while other fields are untouched. Finally, the packet is forwarded to the SINC Egress Proxy.
When the packet reaches the SINC Egress Proxy, it looks up the SPI & SI tables and realizes it is the egress. It removes the NSH encapsulation and forwards the inner packet to the final destination.

SINC over MPLS Considerations In this section, MPLS, which is a layer 2.5 protocol, is used as a running example on how to create a tunnel and encapsulate the SINC header, in order to implement the desired in-network computation. As shown in the , the overall architecture is similar to the SFC solution. In this case, the SINC proxy is also a Label Edge Router. The SW/Rs connecting the SINC proxies and SINC SW/Rs are Label Switching Routers (LSR). Before sending out a SINC packet, the Label Switched Paths (LSP) should be established between the SINC proxies and SINC SW/Rs. The SINC SW/Rs and proxies can identify a SINC packet by the LSPs used. Upon receiving a packet with a SINC header, the SINC Ingress Proxy encapsulates the packet with a MPLS label(s) according to LSP, before forwarding it to the SINC SW/R. The SINC SW/R pops up/pushes the MPLS label before/after the data computation. The results will be forwarded to the SINC Egress Proxy where the MPLS label will be popped up again before the packet is delivered to the destination.

SINC over MPLS Architecture. | LSR |-->| SW/R |-->| LSR |-->| Proxy | +------------+ +-----+ +-----------+ +-----+ +-----------+ ]]>

SINC MPLS encapsulation As shown in the , one or more MPLS labels are added in front of the SINC header.

SINC MPLS Header.

SINC over MPLS Workflow A simple example with one sender (Host A) and one receiver (Host B) is used to illustrate the workflow. Packet processing goes through the following steps:

Host A transmits a packet containing a SINC header together with data to the SINC Ingress Proxy.
Based on the LSP information obtained from control plane, the SINC Ingress Proxy builds up a SINC MPLS header pre-pended to the original packet. Then, according to the LSP information, the SINC packet is forwarded to the next hop.
The LSR forwards the packet based on the LSP information obtained from control plane.
As shown in , when the packet reaches the SINC node, the LSP tunnel ID indicates that this packet contains a SINC header. The SINC SW/R pops up the label and hands the packet to in-network computing module. It is worth noting that the penultimate hop popping must be disabled. Otherwise, an additional mechanism is needed to signal to the SINC node that there is actually a SINC header in the packet.
The in-network computing module verifies the Group ID and Data Source ID in the SINC header, then preforms the required computation defined in the Data Operation field. When it is done, the payload is replaced with the result. The Operation Done flag will be set to 1. The SINC SW/R pushes a MPLS label to the packet. Finally, the packet is forwarded to the SINC egress proxy.
When the packet reaches the SINC Egress Proxy, it looks up the LSP information and realizes it is the egress. It pops up the MPLS label, replaces the innner SINC header with the outer SINC header, and forwards the inner packet to the final destination (Host B).

Security Considerations In-network computing exposes computing data to network devices, which inevitably raises security and privacy considerations. The security problems faced by in-network computing include, but are not limited to:

Trustworthiness of participating devices
Data hijacking and tampering
Private data exposure

This documents assume that the deployment is done in a trusted environment. For example, in a data center network or a private network. A fine security analysis will be provided in future revisions of this memo.

IANA Considerations This memo does not contain any request to IANA.

Acknowledgements Dirk Trossen's feedbacks were of great help in improving this document.