RESEARCH RECORD: ASYMMETRIC OBSERVATION IN DISTRIBUTED PACKET-SWITCHED NETWORKS

Abstract

The Internet is a distributed system of spatially separated processes that communicate by exchanging independent datagrams across interconnected networks. Observation is asymmetric because each node maintains a private local state and receives delayed messages that establish only a partial ordering of events. Measurement is constrained by Autonomous System business relationships and administrative policies that filter or rate-limit diagnostic traffic at the router control plane. Visibility remains fragmented due to finite measurement vantage points and the practical impossibility of collecting state from every network.

Definitions

• Happened-Before (→): A partial ordering in which event a precedes b if they occur in the same process, or if a is the sending and b is the receipt of a message.

• Concurrent: Two events are concurrent if neither can causally affect the other.

• Autonomous System (AS): An independently operated network that participates in interdomain routing under locally defined policy and external business agreements.

• Control Plane: The router subsystem responsible for processing control traffic and generating responses such as TTL-expiry errors.

System Model

Processes are physically distinct and have no knowledge of other processes’ local state until a message is received. Message delay is non-negligible relative to the time between local events, preventing system-wide simultaneity as a primitive.

Packet-switched communication segments information into independently routed datagrams. Delivery is best-effort: datagrams may be discarded without a corresponding report.

Interdomain routing is policy-driven. Path selection is constrained by local policy and business agreements that are not encoded in packet headers and are not globally observable from a single vantage point.

Asymmetric Observation

Node-level visibility is restricted to local event sequences and received messages. The global network cannot be observed directly as a unified object from any single observer.

Topology and performance observations are vantage-point dependent. Measurements reflect a specific source-to-destination path, not the space of possible paths. Alternate routes and contemporaneous routing decisions remain unobserved unless separately measured.

Temporal inconsistency arises when external causality exists outside the system’s observable message graph. Events that are causally related in the external world may appear concurrent or reversed under partial ordering and delay.

Operational opacity follows from administrative controls. Policies such as control-plane rate limiting can suppress diagnostic responses while leaving forwarding behavior unaffected, producing measurements that do not reflect data-plane conditions.

Measurement Limits

Protocol filtering and selective response behavior limit hop-by-hop visibility. Missing responses do not imply the absence of forwarding or the absence of a router on-path.

Inferences about interdomain relationships are underconstrained from the observer’s perspective. Business relationships and routing policy are not directly observable in packet transit and require indirect reconstruction from limited data.

Global reconstruction is resource-bounded. Complete topology capture would require continuous measurement from every network and every relevant vantage point, which is operationally infeasible.

Diagnostics are path-specific by construction. Tools reveal identities and round-trip characteristics along one observed route and provide no guarantees about representativeness across time, traffic classes, or alternate paths.

Core Finding

Network observation is inherently asymmetric and cannot be global because message delay, partial event ordering, and policy-driven opacity prevent any single vantage point from reconciling the disparate local states of all participants.