Appendix C - Glossary¶

This glossary explains the terms used throughout the course - TRex, DPDK and TRex-EMU, InfluxDB and Grafana, the Comware and EVPN-VXLAN mechanisms of the LVR fabric, measurement vocabulary and the names that are specific to the LVR platform. Each entry gives a short definition and points to the chapter where the term is explained in depth. Device names, VLANs and addresses refer to the LVR design described in Part II; general networking terms that every network engineer knows are not repeated here.

16.0.0.0 / 48.0.0.0 (TRex example addresses): The client and server ranges of almost all TRex examples, made for a bench with one device under test whose static routes point 16.0.0.0/8 to the client-side TRex port and 48.0.0.0/8 to the server-side port. They are registered public addresses (HPE, Prudential), so the LVR platform uses reserved pools in the real VLANs instead; the counterpart in LVR stateful tests is 10.9.92.0/24 / 10.9.93.0/24. See Chapter 4.
802.1ad (Provider Bridges): IEEE amendment that introduced the service VLAN tag in front of the customer tag, i.e. two stacked 802.1Q tags (QinQ); it is cited by the 5950 QinQ chapter and is now part of IEEE 802.1Q. On the TX switches both tags use TPID 0x8100, the Comware default. See Chapter 12.
Access mode (VSI): Defines how a VTEP treats the 802.1Q tag of frames on an Ethernet service instance. In the default VLAN access mode the tag is removed when a frame enters the VSI and added again towards the local port, so VXLAN packets carry untagged inner frames. See Chapter 10.
Acquire: A TRex client takes ownership of ports before it may change them; other clients can still read statistics. A forced acquire (-f, reset) takes the ports away from their current owner. See Chapter 4.
Alert rule: A Grafana query plus condition that is evaluated periodically; it fires after the pending period and is routed by notification policies to contact points. See Chapter 9 and Chapter 19.
AP (attachment point): One 10G link from a TX switch into a Leaf or Border Leaf, for example K-L11 or C-BL. On the TX switch each AP is an isolated S-VLAN. See Chapter 13.
AP index: Number assigned to each AP: 1-13 for the Leaf and Border Leaf APs, 14 for the AP FW that exists only with firewall-zone option A. It determines the monitoring client IP (pool start + index - 1) and the fourth byte of the monitoring client MAC; a lab client takes only its IP from the pinned AP (lab pool start .151 + index - 1, i.e. index + 50 relative to the monitoring pool) - its MAC is the fixed lab-port MAC. See Chapter 13.
ASTF (advanced stateful mode): TRex mode in which a TCP/UDP stack per flow emulates clients and servers, so that stateful devices such as firewalls see complete sessions. It runs on trex-lab (unit trex-lab-astf). See Chapter 7.
Average latency (TRex): The average value of a latency stream is low-pass filtered: new average = previous average / 2 + average of the last sampling period / 2. It reacts to changes with a delay. See Chapter 5.
Baseline: The latency and loss values measured in the first days of operation; alert thresholds are derived from it. See Chapter 21.
BE reference (DSCP class): The DSCP 0 (best effort, TOS 0x00) copy in each DSCP group. EF, AF41 and CS1 are compared with this stream, which is identical except for the TOS byte (same pair, size and rate). See Chapter 14.
BL (Border Leaf): sw-rzk-BL and sw-rzc-BL: the VTEPs with the switch gateways (VPN-CORE gateways on sw-rzk-BL), the DCI link and the firewall attachments. The Leafs are Layer 2 only. See Chapter 10.
BPDU drop: Comware bpdu-drop any discards all BPDUs received on a port. It is optional on the TX switch AP ports, if bench tests show that Leaf BPDUs reaching the TRex ports are a problem. See Chapter 12.
Bucket: InfluxDB container for time-series data with a retention period. The platform uses trex (raw data, 90 days) and trex_longterm (1-minute aggregates, 730 days). See Chapter 8.
Budget (guard rails): Limits for the permanent monitoring: at most 120 latency streams (TRex limit 128), 600 flow-stat streams (limit 1023), 50 kpps in total and 100 pps per stream through a firewall. See Chapter 16.
C-VLAN: Customer VLAN: the inner 802.1Q tag. In this platform it is the production VLAN, for example 1314. See Chapter 12.
Capture (TRex): In-memory capture of received or transmitted packets on selected TRex ports with a BPF filter, shown live or written to a PCAP file; it requires service mode. See Chapter 25.
Centralised gateway: EVPN design in which the gateways of a VSI exist on one VTEP only - here the VPN-CORE gateways on sw-rzk-BL. ARP requests from clients behind Leaf APs reach the gateway through VXLAN tunnels; the HPE EVPN guide ties this to ARP flood suppression on the VTEPs, so the resolution from a Leaf AP is a commissioning check. See Chapter 6 and Chapter 21.
Class (traffic class): One of L2, L3-BL, L3-DCFW, L3-ENTFW, DSCP and LAB. The class defines the path a stream takes and its UDP destination port (32000 + class). See Chapter 14.
ClusterXL: Check Point cluster technology. The configurations suggest that CPF-ZV01 (RZK) and CPF-ZV02 (RZC) form one stretched cluster (single OSPF neighbour 10.9.0.22); this must be confirmed by the firewall team. In February 2025 the clusters answered ARP with member MACs, so streams sent to a firewall MAC must be rebuilt after a failover (by the gateway-MAC tracker) unless a virtual MAC is used. See Chapter 15.
Console (trex-console): Interactive command-line client of a TRex server; connects to the RPC and publisher ports and, with --emu, to the EMU server. See Chapter 4.
Convergence test: Measurement during planned maintenance: selected monitoring streams run at 1000 pps (1 ms resolution) and the outage per stream is computed from the lost packets. See Chapter 23.
CPS (connections per second): Main load parameter of ASTF tests; the ASTF multiplier scales it. See Chapter 7.
Data source (Grafana): Connection definition from Grafana to a database. The platform provisions two InfluxDB data sources with Flux, one for trex and one for trex_longterm. See Chapter 9.
DC-FW: The DC firewall cluster CPF-ZV05/ZV06: default gateway of most server VLANs, attached to both Border Leafs via BAGG31 "Vlan-all". Class L3-DCFW measures routing through it. See Chapter 10.
DCI (data-centre interconnect): Route-Aggregation1 between sw-rzk-BL and sw-rzc-BL: 2 x 40G with MACsec and 40G ER4 optics. An inter-DC L2 stream crosses it once per direction. See Chapter 10.
Downsampling: Aggregating raw data into coarser intervals in another bucket. A Flux task writes 1-minute aggregates from trex into trex_longterm. See Chapter 8.
DPDK (Data Plane Development Kit): Libraries and poll-mode drivers that let a user-space program such as TRex read and write NIC rings directly, without interrupts and kernel copies. See Chapter 3.
dpdk_setup_ports.py: TRex helper script: shows the NICs and their drivers (-s, -t), creates configuration files (-i, -c) and, when TRex starts, binds the ports to the DPDK driver and checks hugepages. See Chapter 17.
DSCP class: Copies of an L2 and an L3-ENTFW stream pair (groups dscp-l2-interdc and dscp-entfw) with TOS BE (0x00, the reference), EF (0xB8), AF41 (0x88) and CS1 (0x20), 256 bytes, both directions - 16 latency streams that detect re-marking and queuing differences. On the wire TRex sets bit 0 of the TOS byte (0xB8 leaves as 0xB9); the DSCP is unchanged. See Chapter 14.
dual_if: TRex ports work in pairs (ports 0 and 1, ports 2 and 3). The platform section of a TRex configuration file has one dual_if entry per port pair with its NUMA socket and data-plane threads. See Chapter 3.
EMU (TRex-EMU): Emulation server written in Go, started with t-rex-64 --emu together with --software (RPC port 4510, one EMU server per host). It emulates hosts that answer ARP and ICMP and resolve their gateway; it exchanges packets with the TRex RX core over ZMQ. Loading an EMU profile first removes all existing namespaces and clients. See Chapter 6.
EMU client: One emulated host with MAC address, IPv4 address, default gateway and plugins such as arp and icmp. The arp plugin queries the default gateway again every timer seconds (60 by default). See Chapter 6.
EMU namespace: Group of EMU clients identified by {port, two VLAN tags, two TPIDs}; tci[0] is the outer tag. Clients in one namespace share the ARP cache, so a gateway is resolved once. See Chapter 6.
ENT-FW: The Enterprise firewall cluster CPF-ZV01/ZV02, hub between the VRFs; OSPF neighbour 10.9.0.22 of VPN-CORE on Vsi40. Class L3-ENTFW measures the path through it. See Chapter 10.
Ethernet service instance: Comware object on a Layer 2 port (service-instance N / encapsulation s-vid N) that matches a VLAN and maps it to a VSI with xconnect vsi. Every server port and every AP port of the fabric uses it. See Chapter 10.
EVPN: BGP control plane for VXLAN that distributes MAC and IP reachability (RFC 7432, RFC 8365). In the fabric the spines are route reflectors and the Border Leafs exchange EVPN routes across the DCI. See Chapter 10.
Exporter: Planned trexctl component that polls the TRex statistics every 2 seconds, computes interval deltas and writes them to InfluxDB. It is a read-only client (never acquires the trex-mon ports) and also runs the supervisor thread. See Chapter 18.
Fibre offset: Extra delay of the TX-to-TX fibre (about 5 us per km). Every frame starts and ends at the server in RZK, so an RZK-RZC stream crosses that fibre once and an RZC-RZC stream twice. See Chapter 14.
Field (InfluxDB): Measured value of a point, for example rx_pps or lat_max_us. Fields are not indexed. See Chapter 8.
Filtered service mode: Service mode in which only selected packet types (mask) go to the service path, while traffic and statistics keep running; service --emu uses it for EMU. See Chapter 6.
Firewall test zone: Dedicated TRex subnet 10.9.90.0/24 behind the Enterprise firewall (VIP 10.9.90.1), the destination of class L3-ENTFW. The plan recommends option B and uses it as the default: the zone is VLAN 4003, reached through the Border Leaf APs, with the zone clients 10.9.90.106 (K-BL) and 10.9.90.113 (C-BL) and the lab zone client 10.9.90.156. With option A, the alternative, it would be VLAN 1420 on new firewall ports, zone client 10.9.90.114 (AP FW, index 14). The choice is still an open decision. See Chapter 15.
Flow-stat stream: Stream with STLFlowStats(pg_id): TRex counts its transmitted and received packets, which gives loss but no latency. Up to 1023 per instance with --software, 127 on X710 in hardware mode. See Chapter 5.
Flux: Functional query and scripting language of InfluxDB 2.x, used by Grafana panels and the downsampling task. See Chapter 8.
FWZ: Inventory alias of the firewall zone with option B: FWZ: { alias_of: K-BL, vlan: 4003 }, i.e. the AP K-BL with VLAN 4003. Lab port p6 is pinned to it by default (TX switch port XGE1/1/11, PVID 3110), and test commands address the zone as --dst FWZ --dst-vlan 4003. See Chapter 18.
gateway_mac_changed: Kind of trex_event written by the gateway-MAC tracker after it has rebuilt streams for a new gateway MAC; the Grafana alert "Gateway MAC changed" (severity info) fires on it. See Gateway-MAC tracker.
Gateway-MAC tracker: Planned trexctl function for every VLAN whose gateway is a firewall cluster (gateway_mac_tracking: true: the zone VLAN 4003 - 1420 with option A - and the production VLANs). It polls the gateway MAC that EMU has resolved (dgw.rmac) every 10 s and forces a re-resolution every 60 s; when the MAC differs from the one in the running streams, it rebuilds the affected streams (same pg_ids, counters restart) and writes gateway_mac_changed. How quickly EMU learns the new MAC depends on how ClusterXL announces it, so the plan's 15-s recovery target is still to be confirmed. See Chapter 18 and Chapter 23.
Gratuitous ARP: ARP message announcing the sender's own address; EMU uses it so that the fabric learns a client MAC before traffic arrives. In the other direction, EMU updates its entry for a sender at once when it receives an ARP request, gratuitous requests included; an ARP reply sent to the broadcast address is ignored. See Chapter 6.
Hardware mode: TRex mode without --software: on X710 the NIC counts flow statistics (127 pg_ids, at most one VLAN tag, packet counters only). Used by trex-lab for line-rate tests. See Chapter 5.
Histogram (latency): Per latency stream, counts of packets per logarithmic latency bucket; the key is the lower bucket bound in microseconds. Differences between two readings give the distribution of an interval. See Chapter 5.
Hugepages: Large memory pages (2 MB on this server) that DPDK uses for packet buffers; 4096 pages (8 GB) are reserved on the kernel command line for both TRex instances. See Chapter 3.
hw_id: Internal index that TRex assigns to the pg_id of a statistics stream. It is what the packets carry - in the IPv4 ID of flow-stat streams and in the latency signature - while the user-facing pg_id stays in the TRex statistics. See Chapter 5.
Interval maximum: Highest latency of one exporter interval (field lat_max_us), derived from histogram differences, because last_max covers only 0.5 s; the field lat_last_max_us keeps last_max for reference only. See Chapter 16.
IP ID (statistics streams): TRex rewrites the IPv4 ID of statistics streams. The documentation states the reserved range 0xff00-0xffff; the v3.x source uses 0xFC00 + hw_id for flow-stat streams and 0xFFFF for every latency stream - the value is derived from the internal hw_id, not from the pg_id. See Chapter 5.
IRF (Intelligent Resilient Framework): Comware stacking: two switches act as one device. Each AP uses a port on one member, so a member reboot takes that AP down. See Chapter 10.
Jitter: Variation of the delay. TRex computes it per stream as described in RFC 3550 appendix A.8. See Chapter 1.
kind (latency / flowstat): Attribute of a monitoring stream group and InfluxDB tag that says whether the streams are latency streams or loss-only flow-stat streams. See Chapter 16.
L2 (class): Streams between two APs in the same VLAN; the destination MAC is the remote TRex client. Intra-DC and inter-DC pairs. See Chapter 14.
L3-BL (class): VLAN 1314 to VLAN 1315, routed by the gateways on sw-rzk-BL in VPN-CORE; a routed stream between two RZC APs crosses the DCI twice (to Kalk and back), between an RZC and an RZK AP once. See Chapter 14.
L3-DCFW (class): Between two production VLANs whose default gateway is the DC firewall; needs DC-FW rules. See Chapter 14.
L3-ENTFW (class): VLAN 1314 in VPN-CORE through sw-rzk-BL, Vsi40 and the Enterprise firewall into the TRex firewall zone - with option B, the plan's default, VLAN 4003, back through BAGG10 to the zone client at the Border Leaf AP of the source's data centre. See Chapter 14.
LAB (class): On-demand tests on trex-lab: throughput (no-drop rate), soak, convergence and ASTF; UDP port 32009. See Chapter 22.
Lab port (pinned port): TX switch port of trex-lab (p4-p7) configured as QinQ access port with the S-VLAN of one AP; TRex sends single-tagged frames. Default pins: p4 K-L11 (3101), p5 K-L41 (3105), p6 FWZ = K-BL (3110), p7 C-L21 (3204). Re-pinning changes port access vlan; the lab-port MAC stays, only the lab IP follows the new AP. See Chapter 12.
Lab-port MAC: The fixed MAC of a lab port, set as src_mac in the port_info of trex_lab.yaml and trex_lab_astf.yaml and recorded as mac: in the inventory: p4 02:54:52:f0:00:04, p5 :05, p6 :06, p7 :07 (Comware notation 0254-52f0-0004 ...). TRex answers ARP for a lab port and runs ASTF with the port MAC, and a lab port is re-pinned over time, so the MAC cannot depend on the AP; only the lab IP does (lab pool start .151 + AP index - 1). The fixed MAC keeps the static MAC entries on the TX switch and the firewall objects stable across re-pins. The monitoring clients (EMU) keep the per-AP formula 02:54:52:<index>:<VLAN hex>. See Chapter 13 and Appendix B.1.
last_max: Highest latency of a stream in the last 0.5 s sampling window only; reading it every 2 s misses peaks. See Chapter 5.
Latency stream: Stream with STLFlowLatencyStats(pg_id): TRex writes a 16-byte signature (internal hw_id, sequence number, TX timestamp) at the end of the payload; the RX core computes latency, jitter and the sequence error counters (dropped, dup, out_of_order, seq_too_high, seq_too_low). At most 128 per instance; built with UDP checksum 0, because the signature is written into every packet without recalculating the checksum. The multiplier does not apply: a new rate means removing and re-adding the stream. See Chapter 5.
Line protocol: Text format in which points are written to InfluxDB: measurement, tags, fields, timestamp. See Chapter 8.
Loss resolution: Shortest outage a stream can show: 1 / pps, for example 5 ms at 200 pps. See Chapter 1.
Master thread: master_thread_id: the TRex control thread (RPC server, statistics); TRex places the helper services on it by default. See Chapter 3.
Measurement (InfluxDB): Name grouping related points, comparable to a table; the platform plans trex_stream, trex_port, trex_global, trex_test and trex_event. See Chapter 8.
Monitoring trunk (multiplexed port): TX switch port of trex-mon (p0-p3): an 802.1Q trunk permitting all S-VLANs, so frames reach TRex double-tagged and every AP is reachable from every monitoring port. See Chapter 12.
Multiplier: Rate factor of start and update. STL accepts rates (-m 1kpps, -m 10mbps, -m 50%) or a plain factor; ASTF -m is a plain multiplier of the connection rate. It does not change latency streams. See Chapter 4.
Mute timing: Recurring Grafana time interval in which notifications are suppressed, for example a maintenance window; alert evaluation continues. See Chapter 9.
NDR (no-drop rate): Highest offered load without packet loss, found by a binary search as in the RFC 2544 throughput test. See Chapter 22.
NUMA: Non-uniform memory access: memory and PCI devices belong to a CPU socket. NICs, hugepages and TRex threads should be on the same node. See Chapter 3.
One-way latency: Delay from sender to receiver in one direction. TRex can measure it because both ends of every stream are on the same server and share one clock. See Chapter 1.
ONE_QUE / MULTI_QUE: Queue modes of TRex, see Queue mode.
Option A / Option B: The two designs for attaching the firewall test zone. B, recommended by the plan and its default: the existing, free transfer VSI 4003, already trunked on BAGG10 to both firewall members and reached through the Border Leaf APs K-BL and C-BL. A, the alternative and the customer's original wish: new physical firewall interfaces cabled to the TX switches (S-VLAN 3300, VLAN 1420), with the risk that ClusterXL fails over because of the non-redundant test path. The TX switches already carry the option-A ports XGE1/1/7 (rzk) and XGE1/1/8 (rzc) configured but shut down, and S-VLAN 3300 is defined and permitted on the trunks without an active member port, so A later needs only cabling and undo shutdown. The choice is still open. See Chapter 15.
Outage duration: Lost packets of a stream divided by its rate in packets per second. See Chapter 23.
pg_id (packet group ID): 32-bit identifier of a stream's statistics; each latency stream needs its own. Planned ranges: latency 1-127, flow-stat 1000-1999, lab 2000 and above; trexctl keeps the allocation stable in /data/trexctl/state/pgid-map.json. Inside the packets TRex uses an internal hw_id instead. See Chapter 5.
Port number (instance-local): TRex numbers the ports of each instance from 0 in the order of its interfaces list. trex-mon: ports 0-3 = p0-p3; trex-lab: ports 0-3 = p4-p7; trex-lab-astf: port 0 = p4 (client side), port 1 = p6 (server side). Console and API always use these numbers. See Chapter 3.
prefix: Name of a TRex instance's DPDK runtime and hugepage files; mon and lab keep the two instances apart. See Chapter 3.
Profile (stateless): Python file whose get_streams() returns the streams to load; started with start -f. Tunables pass parameters. See Chapter 4.
Profile (trexctl): Planned YAML file (for example monitoring-default.yaml) that describes stream groups by kind, class, VLANs, pairs, rate and size. See Chapter 18.
Promiscuous mode: Port attribute with which a TRex port accepts frames for any destination MAC; needed because EMU client MACs are not the port MAC. See Chapter 6.
Provisioning (Grafana): Creating data sources, dashboards and alert rules from files at start-up, so the installation is reproducible. See Chapter 9.
PVID: Port VLAN ID. On a QinQ-enabled TX switch port the PVID is pushed as outer tag on every incoming frame; it is the AP's S-VLAN. See Chapter 12.
QinQ: Two stacked 802.1Q tags. Comware qinq enable adds the port PVID as outer tag to all incoming frames and removes it on egress. See Chapter 12.
Queue mode: How TRex distributes received packets, chosen at start-up and printed as "set dpdk queues mode to ...". ONE_QUE (software mode, c: 1): the RX core reads each port's receive queue itself - accurate latency; used by trex-mon. MULTI_QUE (software mode, c >= 2): data-plane threads read the queues and pass latency packets on - latency not accurate. DROP_QUE_FILTER (hardware mode on X710): a NIC filter steers latency packets to the RX core - accurate with any c. See Chapter 3.
RFC 2544: IETF benchmarking methodology for network interconnect devices; its throughput test is the basis of the no-drop rate search. See Chapter 22.
RX core: latency_thread_id: the TRex thread that processes received latency and flow-stat packets in software and the service traffic; it busy-polls while latency streams are active. In ONE_QUE mode it reads the port receive queues itself. See Chapter 3.
S-VLAN: Service VLAN: the outer tag. On the TX switches it identifies the AP (3101-3110 RZK, 3201-3210 RZC; 3300 is the stretched zone S-VLAN of option A - defined and permitted on the trunks of both TX switches, but without an active member port with option B). See Chapter 12.
Scapy: Python packet library with which TRex profiles describe packets (Ether()/Dot1Q()/IP()/UDP()). See Chapter 4.
Scapy server: TRex helper daemon on port 4507, one per host. Every interactive stateless start of t-rex-64 restarts it unless --no-scapy-server is given, so the lab units run with that flag and never restart the helper of trex-mon. See Chapter 3 and Chapter 17.
Service mode: Port mode in which received packets go to the service path, so the port can answer ARP and ping and run captures. See Chapter 6.
services_core: CPU on which the TRex helper services (Scapy server, EMU) are pinned; default is the master thread. trex-mon sets it to its master thread 4, not to the busy-polling RX thread 5. See Chapter 3.
Software mode (--software): TRex mode without hardware filters: all received packets are processed by software. Required for QinQ flow statistics and for 1023 flow-stat streams; used by trex-mon with c: 1, so that the queue mode is ONE_QUE. See Chapter 5.
STL (stateless mode): TRex mode in which streams are generated packet by packet without TCP state; used for all monitoring and throughput tests. See Chapter 4.
Stream: One unidirectional TRex packet flow with its own rate, packet template and statistics (pg_id). See Chapter 4.
Supervisor thread: Planned thread of the exporter that watches trex-mon: when it detects a restart (server uptime reset, or active_streams falling to zero while a profile is recorded as active in /data/trexctl/state/), it runs trexctl apply for the active profile as soon as the server accepts connections again and restarts the gateway-MAC tracker. It works because trexctl apply leaves the streams running without an owner (disconnect(stop_traffic=False, release_ports=True)) and re-acquires the ports with force=True. See Chapter 18 and Chapter 20.
Tag (InfluxDB): Indexed key-value pair of a point (for example class, src_ap, dc_pair); Grafana filters and groups by tags. See Chapter 8.
Task (InfluxDB): Flux script that InfluxDB runs on a schedule (every, offset), for example the downsampling task. See Chapter 8.
Token (InfluxDB API token): Credential with read or write permissions on buckets; since InfluxDB 2.9 tokens are stored hashed and shown only when created. See Chapter 8.
TPID: Tag protocol identifier, the EtherType of a VLAN tag; 0x8100 by default for both tags on the TX switches and in EMU. See Chapter 12.
Transfer VSI: VSI that connects a VRF to a firewall, for example 4002 "Transfer VPN-CORE ENT-FW" (Vsi40, 10.9.0.16/29) or the free VSI 4003, which becomes the firewall test zone with option B, the plan's default. See Chapter 10.
trex-cfg: Script that t-rex-64 runs before every start: it binds the ports, checks hugepages and starts the helper daemons; on Red Hat-family systems it also stops firewalld. Host protection of the server is therefore a design decision. See Chapter 17.
trex-lab: TRex instance for on-demand tests on NIC2 (p4-p7, TRex ports 0-3): stateless hardware mode or ASTF, never both at once. The ASTF unit uses its own configuration with p4 (port 0, client side) and p6 (port 1, server side). Both units run with --no-scapy-server, the ASTF unit also with --unbind-unused-ports, and never with --emu. See Chapter 11.
trex-mon: TRex instance for permanent monitoring on NIC1 (p0-p3): stateless, --software --emu with c: 1, double-tagged frames. See Chapter 11.
trexctl: Planned orchestrator (not yet developed) that turns the inventory and profiles into EMU clients and streams, runs tests and exports statistics. Until it exists, the native console and API are used. See Chapter 18.
TUI: Text user interface of the TRex console (tui) with live dashboard, stream, latency and utilisation views. Like stats -s and stats -l, the stream view shows at most 4 pg_ids and the latency view at most 5. See Chapter 20.
TX switch: Dedicated TRex distribution switch per data centre (sw-rzk-TX1, sw-rzc-TX1, spare 5950s): QinQ only, no STP, no production VLANs, no routing. See Chapter 12.
TX-to-TX link: Dedicated dark fibre or DWDM channel between the two TX switches; carries S-VLANs 3201-3210 and the stretched zone S-VLAN 3300 (harmless with option B, where 3300 has no member port; required with firewall-zone option A), so the RZC APs reach the server in RZK. See Chapter 11.
Unknown-unicast flooding: A switch sends a frame with an unlearned destination MAC to all ports of the VLAN. For lab tests this would reach the monitoring trunks and the production VSI; static MAC entries and pre-learning prevent it. See Chapter 24.
Variable (Grafana): Dashboard placeholder such as ${class} or ${dc_pair:regex} that the user selects and Grafana inserts into queries. See Chapter 9.
vfio-pci: Linux driver that gives a user-space program safe access to a PCI device through the IOMMU; the preferred DPDK driver because it works with Secure Boot. See Chapter 17.
VLAN 1314 / 1315: Existing "Performance-Test" VLANs (VSI VPN-1314/1315, gateways 10.3.14.1 and 10.3.15.1 on sw-rzk-BL in VPN-CORE), the main carriers of the monitoring. They are also mapped towards the DC firewall on both Border Leafs. See Chapter 10.
VLAN 4003 / VSI VPN-4003: Transfer VSI "Transfer ENTFW *frei*" (vxlan 4003, no gateway interface on the Border Leafs), trunked on BAGG10 to both Enterprise firewall members and unused in February 2025. With firewall-zone option B, the plan's default, it is the TRex firewall zone: the firewall adds a VLAN 4003 interface with the VIP 10.9.90.1, and TRex reaches it through service instance 4003 on the K-BL and C-BL AP ports. See Chapter 15.
VPN-CORE: The main VRF (vpn-instance) of the fabric; its gateways are on sw-rzk-BL and its default route points to the Enterprise firewall. See Chapter 10.
VSI (virtual switch instance): Comware Layer 2 domain bound to a VXLAN VNI, named VPN-<vlan> in this fabric. See Chapter 10.
Vsi-interface: Layer 3 interface of a VSI, i.e. a gateway, for example Vsi-interface914 10.3.14.1/24 on sw-rzk-BL. See Chapter 10.
VTEP: VXLAN tunnel endpoint: a Leaf or Border Leaf that encapsulates frames into VXLAN and decapsulates them. See Chapter 10.
VXLAN: Layer 2 frames carried over UDP/IP between VTEPs (RFC 7348). See Chapter 10.
X710: Intel 10G NIC family (X710-DA4: 4 x SFP+), the preferred TRex NIC: hardware flow statistics, Intel-coded optics only (passive DAC usually works). See Chapter 3.
ZMQ (ZeroMQ): Messaging library that TRex uses for its RPC channel (4501), its publisher (4500) and the packet channel to EMU. See Chapter 2.