Concise Encoding of Flow Attributes in SDN Switches

Abstract

Network devices such as routers and switches forward traffic based on entries in their local forwarding tables. Although these forwarding tables conventionally make decisions based on a packet header field such as a destination address, tagging flows with sets or sequences of attributes and making forwarding decisions based on these attributes can enable richer network policies. For example, devices at the edge of a network could add a tag to each packet that encodes a set of egress locations, a set of host permissions, or a sequence of middleboxes to traverse; simpler devices in the core of the network could then forward packets based on this tag. Unfortunately, naive construction of these tags can create forwarding tables that grow quadratically with the number of elements in the set or sequence-prohibitive for commodity network devices. In this paper, we present PathSets, a compression algorithm that makes such encodings practical. The algorithm encodes sets or sequences (e.g., middlebox service chains, lists of next-hop network devices) in a compact tag that fits in a small packet-header field. Our evaluation shows that PathSets can encode attribute sets and sequences for large networks using tag widths competitive with existing approaches and that the number of forwarding rules grows linearly with the number of attributes encoded.

Why (and How) Networks Should Run Themselves

“…The ability to tag packets with state and update that state at subsequent hops enables the data plane to support a range of powerful abstractions. For example, a packet header could carry the version of network policy applied to that packet (e.g., to support consistent policy updates [25], sets or sequences (e.g., of next-hops, network paths, or middleboxes for flexible traffic steering [21]); states of a deterministic finite automaton to evaluate a regular expression on the properties of a packet and its path through the network, to measure or control traffic based on these properties [23]; or the aggregation of traffic statistics across a path, to collect path-level metrics such as maximum link utilization or total queuing delay [13].…”

RDNA: Residue-Defined Networking Architecture Enabling Ultra-Reliable Low-Latency Datacenters

“…In literature there have been numerous proposals that make packet forwarding based on labels including Multiprotocol Label Switching (MPLS) [9], Virtual Local Area Network (VLAN), KeyFlow [16], and SDN-based architectures [10], [11], [12]. MPLS is a well-known source-routing protocol that forwards packets by writing and matching on labels attached to packets.…”

“…In this paper, we propose RDNA (Residue-Defined Networking Architecture), an approach that explores the Residues Number System (RNS) [8] as a fundamental concept to facilitate and enhance network programmability. Differently from other works in the literature, such as [9], [10], [11], [12], RDNA main contribution involves a minimalist forwarding model for the core nodes. Instead of forwarding packets based on traditional table lookup operations, the core nodes are tableless switches that forward packets using merely remainder of the division (modulo) operations.…”

Programmable residues defined networks for edge data centres

“…RELATED WORK There have been many proposals in the literature to make forwarding decisions based on flat tags or labels including MPLS [19], VLANs [20] and SDN-based architectures [21], [22], [23]. MPLS [19] is a well-known source-routing protocol that forwards packets by writing and matching on labels attached to packets.…”