Computer Networks: Lecture Course (Rolik, 2016–2017)

This archive preserves twelve lecture decks written by Anatolii I. Rolik for KPI's computer-networks course. The combined course is 1,392 pages and moves from architectural models through medium access and Ethernet to bridged networks and troubleshooting. Two image-only decks were OCRed, and every page was rendered for the visual audit.

The course is an admissible instructor-authored secondary source, not a current standards authority. It is strongest when it explains why layers exist, how a learning bridge builds forwarding state, why loops destabilize a bridged LAN, and how a troubleshooter narrows a fault domain. GTK uses current IEEE, IETF, Linux, and Wireshark primary sources for wire semantics and operational practice.

Complete folder audit #

The two supplied folders contain 26 files. The source archive contains only the twelve instructor-authored decks from TKSM-2/; no student submission or application metadata was copied into the library.

The student-work exclusion is intentional, not a coverage gap. The durable topics taught by those exercises—VLANs, wireless access, and VPNs—already exist independently in the networking taxonomy and must be written from authoritative sources.

Deck-by-deck concept map #

Page ranges refer to the combined archive.

Mechanisms worth preserving #

A layer is a contract boundary, not a seven-box classification game #

The useful OSI lesson is the separation between a service exposed upward, an interface through which it is requested, and a protocol exchanged between peer entities. Encapsulation follows from that separation: each layer adds the control information needed to realize its service while treating the upper-layer unit as payload. The exact number and names of layers matter less than knowing which contract owns framing, delivery scope, reliability, or naming.

This material was already covered at deep depth in GTK. Creating another OSI chapter would duplicate facts without improving the territory.

Medium access is allocation under incomplete coordination #

The access deck organizes schemes by what resource is divided—time, frequency, code, or space—and by how grants are made—fixed assignment, reservation, polling, contention, or a hybrid. That matrix transfers beyond old bus LANs to radio systems, passive optical access, industrial buses, and scheduled fabrics. GTK's existing shared-medium chapters already generalize it and explain the capacity, delay, fairness, and coordination trade-offs.

Ethernet separates a stable MAC service from changing PHYs #

The course's four Ethernet decks make one durable architectural point: the recognizable MAC frame and service persist while line rate, coding, medium, and media-independent interface change. A learner should therefore reason separately about frame semantics, link mode, and the selected PHY. Memorizing historical suffixes such as 10BASE5 or individual Fast/Gigabit coding tables does not provide that separation.

Transparent bridges learn from sources and forward on destinations #

A bridge observes the source MAC address and ingress forwarding context, refreshes an ageable FDB entry, and then uses the destination MAC address to choose local delivery, a known egress, or flooding. This asymmetry explains both the appeal and the failure modes: learning needs no endpoint signalling, but unknown unicast and group traffic still replicate, host movement temporarily invalidates state, and loops can multiply frames faster than tables converge.

VLANs do not merely decorate frames. They create separate forwarding contexts. An operational FDB lookup is therefore conceptually keyed by both VLAN and destination MAC; ingress classification and egress tag treatment determine which context a frame enters and how it leaves.

Redundant bridged paths require an explicit loop-avoidance mechanism #

Ethernet has no hop limit in its MAC header. A physical cycle can therefore keep a broadcast, multicast, or unknown-unicast frame alive, create multiple copies, and destabilize learning. Spanning-tree protocols solve this by electing a root, assigning path roles, and preventing some ports from carrying ordinary data so the active topology is loop-free. The blocked path is not wasted design intent—it is standby capacity unless a multi-path control plane or link aggregation safely makes it usable.

Troubleshooting is controlled discrimination between hypotheses #

The final deck's ten-step method is most valuable when reduced to an evidence loop: define the symptom precisely, establish the last known good boundary, reproduce without increasing harm, list discriminating hypotheses, observe at the nearest useful boundary, change one cause, and verify both success and non-regression. Packet capture strengthens this loop only when the capture point, direction, clock, snap length, packet loss, and hardware offloads are known.

Current-use boundaries #

• IEEE 802.3 still defines both half-duplex CSMA/CD and full-duplex operation, but hubs, shared coax, collision-domain diameter calculations, and late-collision tuning are historical for ordinary switched Ethernet. They remain useful for recognizing duplex mismatches and legacy equipment, not as a default LAN design.
• The course stops at early Gigabit Ethernet. Current Ethernet spans many copper, optical, backplane, automotive, and access PHYs. GTK teaches the invariant MAC/link model and sends PHY-specific questions to the current clause or manufacturer data rather than extending a memorization catalogue.
• Standalone IEEE 802.1D is no longer the current home of bridging and spanning-tree semantics; these functions are consolidated in IEEE 802.1Q. Historical STP/RSTP names remain useful, but present-day claims cite the current bridging standard.
• Port-based VLAN classification is operationally central. MAC-based and protocol-based membership examples in the course are valid possibilities, but they are not a reliable default design and can make identity, movement, and troubleshooting opaque.
• A VLAN is not a security boundary by itself. It constrains layer-2 forwarding context; policy enforcement still requires explicit admission, filtering, routing, and management-plane controls.
• Link aggregation combines compatible point-to-point links into one logical link and can add resilience, but per-flow hashing does not make one flow use the sum of all member rates. Member ordering, failure detection, partner state, and asymmetric configuration remain operational concerns.
• The centralized-bus deck primarily describes a MIL-STD-1553-style command-response system. Its polling and redundancy ideas generalize; its message layouts, timing, and controller assumptions do not describe contemporary switched enterprise Ethernet.
• The troubleshooting deck predates widespread NIC offload, virtualization, overlays, encrypted transports, high-resolution pcapng metadata, and modern capture privacy practice. GTK retains the hypothesis method and supplies those missing observation constraints.

Visual audit #

All 1,392 pages were rendered and inspected. Доступ.pdf and Trubl_ACTS_KPI.pdf are image-only scans and were OCRed before concept mapping. The final archive was rendered at representative section boundaries, and all twelve merge boundaries were checked against the original page text and geometry.

1. Layer and encapsulation stacks: repeated vertical OSI diagrams clarify peer protocols, service boundaries, and nested PDUs. They are not copied because GTK already has a cleaner, current diagram in the layering chapter.
2. Medium-allocation diagrams: time slots, frequency bands, code separation, polling loops, and contention timelines are load-bearing. Existing GTK shared-medium visuals already preserve the general mechanism without inheriting the source's radio- and bus-specific notation.
3. Ethernet frame drawings: several diagrams mix preamble/SFD, MAC frame fields, and on-medium overhead. GTK redraws one byte-scaled view and explicitly distinguishes the MAC frame from physical-layer timing so frame-size arithmetic is not ambiguous.
4. PHY and cabling illustrations: connector photographs, pinouts, encoding waveforms, repeater topologies, and old media families are mostly variant-specific or historical. Current media and coding entries retain only mechanisms that remain necessary for selection and diagnosis.
5. Learning-bridge and VLAN paths: source/destination learning examples, forwarding tables, tagged links, and per-VLAN topologies are essential. GTK redraws these as a VLAN-scoped FDB state transition and an ingress/egress classification path.
6. Spanning-tree topologies: loop, root-election, port-role, and reconvergence figures carry the causal explanation. GTK redraws a small redundant topology with one alternate path and explains what changes after failure.
7. Switch-fabric diagrams: shared-bus, shared-memory, and crossbar examples explain head-of-line and fabric constraints, but the particular device architectures are dated. They are mapped to the planned queueing and packet-scheduling chapter instead of reproduced here.
8. Troubleshooting flowcharts: the deck's process figures are visually dense but conceptually useful. GTK replaces them with an evidence-at-boundaries diagram that ties each observation to a falsifiable hypothesis.

No source image was copied. Every retained visual is an independent, theme-aware SVG designed to remain legible at article width and checked against current primary sources.