Glossary

base port

7869 (EGO_LIM_PORT), the anchor of the fixed 7869-7875 range. Must be identical on every Symphony host — management and compute, either site. One of the network invariants that costs real debugging days if broken (Network model).

CDI

Containerized Data Importer — the OpenShift Virtualization component that imports/clones VM disk images. Produces the per-VM DataVolume that a VirtualMachinePool clone starts from.

DataSource / DataVolume

A DataSource is a pointer to a "golden" boot source — here, ws2019-symphony, the sysprep-generalized Windows image the golden-image pipeline produces. A DataVolume is the actual per-VM disk CDI clones from that DataSource each time the compute VirtualMachinePool spins up a new host.

Direct Link / Direct Connect / TGW

The production interconnect path: IBM Direct Link Connect ⇔ AWS Direct Connect hosted VIF, stitched by a partner (Megaport/Equinix), routed on the IBM side via Transit Gateway (TGW). The PoC/fallback is IPsec site-to-site VPN (ADR 0004).

dynamic host

A Symphony compute host that joins the grid at runtime rather than being defined statically at install — the join model every Windows compute VM in this build uses. Requires EGO_DYNAMIC_HOST_WAIT_TIME to be set, or the join silently fails.

EGO

Enterprise Grid Orchestrator — the resource-management layer under Symphony (LIM, PEM, vemkd, EGO_CONFDIR config) that every host, master and compute, runs.

egosh

The EGO shell/CLI. egosh resource list is the primary diagnostic for confirming a host actually joined the grid — precisely the check this build’s Windows join is waiting on before "fixed in source" becomes "confirmed" (Architecture, status).

HCO

HyperConverged Operator — the OpenShift Virtualization operator that reconciles KubeVirt, CDI, and the rest of the virt stack from a single HyperConverged CR. Measured convergence time on sym-aws: ~5 min.

Host Factory

Symphony’s Advanced Edition burst mechanism — a pluggable 5-script provider interface (getAvailableTemplates, requestMachines, requestReturnMachines, getRequestStatus, getReturnRequests). The Host Factory KubeVirt provider is this project’s one custom component (detail).

instancetype (d1)

The KubeVirt instancetype family every Symphony compute VM uses — dedicatedCpuPlacement + isolated emulator threads. Required both for license-compliant Windows and for the CPU Manager static policy the virt-metal MachineSet runs (ADR 0013).

IPI

Installer-Provisioned Infrastructure — the OpenShift install method sym-aws actually runs on (self-hosted OCP 4.20 on AWS), adopted after D9 dropped ROSA HCP over an AWS Marketplace billing-link blocker (ADR 0009).

KEDA / CMA

Kubernetes Event-Driven Autoscaling, shipped on OpenShift as the Custom Metrics Autoscaler operator. It can mechanically scale a VirtualMachinePool’s `/scale subresource, but is scoped to a demo only here — scale-in is not grid-aware (ADR 0005).

KubeVirt

The upstream project OpenShift Virtualization packages. VirtualMachine and VirtualMachinePool are its CRDs — the actual API both the Windows compute pool and the Host Factory provider target.

LIM

Load Information Manager — the EGO daemon every host runs. On Windows it installs as a service literally named LIM, not EGO* — a real join-debugging trap found in this build. The master’s LIM dials back to a newly-registered host, so an inbound-only compute-side firewall rule leaves it silently unavail even when outbound checks pass clean.

MachineSet / MCP

A MachineSet is the OpenShift CR that provisions a pool of Machines — here, the virt-metal MachineSet of m5zn.metal bare-metal workers. MachinePool (MCP) is the equivalent construct on managed clusters (ROSA/ROKS-style).

MTU / MSS

Maximum Transmission Unit / Maximum Segment Size. Must be identical end-to-end on the Symphony network — pinned to 1500 on masters and Windows guests. A masters-at-jumbo / guests-at-1500 mismatch is the exact, IBM-documented (KB ibm16164145) root cause behind one real join failure in this build (ADR 0014).

MTV

Migration Toolkit for Virtualization — Red Hat’s VM-migration tool (e.g. from vSphere onto OpenShift Virtualization). Referenced as the standard onboarding path elsewhere in the ecosystem; not used by this greenfield build.

multicluster / SMCP

Symphony’s Advanced Edition federation feature — independent clusters per site, joined via a small primary cluster plus one SMCP proxy per member. The documented fallback topology if the stretched single-cluster WAN design proves fragile (ADR 0010).

OCP Virt

OpenShift Virtualization — the GA Red Hat product (HCO + KubeVirt + CDI) the entire compute plane runs on, chosen over WMCO Windows containers (ADR 0001).

PEM

Process Execution Manager — the EGO daemon that runs and monitors tasks on a host. On Windows compute hosts it is a LIM-spawned child process, not a separate Service Control Manager service.

ROKS

Red Hat OpenShift on IBM Cloud — the designed Phase-3 burst site (VPC bare-metal workers, manual OpenShift Virtualization install). Designed, not yet built.

ROSA

Red Hat OpenShift Service on AWS — the original primary-site platform choice, dropped for self-hosted IPI by D9 after an AWS Marketplace billing-link blocker with no CLI/API workaround. The gitops/clusters/rosa/ directory name is a known cosmetic leftover from before the rename to sym-aws.

SD

Session Director — the Symphony daemon a client reaches via vemkd, which hands off to SSM. SD_ADMIN_PORT/SD_SDK_PORT are 7874/7875 in the fixed base-port block.

SIM

Service Instance Manager — the compute-side counterpart to SSM. The long-lived SSM↔SIM channel is one of the two idle channels this build’s keepalive tuning (EGO_ENABLE_CHAN_KEEPALIVE) protects across the WAN.

SOAM

Service-Oriented Application Middleware — the Symphony session/workload layer (SOA sessions, service packages), installed alongside EGO on every management and compute host.

SSM

Session Scheduler Manager — the management-side session/task-dispatch daemon. Its ports are OS-random by default and must be pinned in sd.xml (symphony/config/sd-ssm-ports.xml) for any firewall or NLB sitting between sites.

symA requestor

The stock Host Factory demand-signal component (throughput policy, tasks/min/ slot) that drives burst requests. The reason a custom KEDA/Prometheus demand signal was rejected: it would duplicate work Host Factory already does natively.

sysprep specialize

The Windows sysprep pass that runs on first boot of each cloned compute VM. This build’s firstboot-symphony.ps1 runs in it — enabling WinRM, opening the EGO port range via netsh advfirewall, rewriting EGO_MASTER_LIST, and starting the LIM service — several steps exist specifically because normal cmdlets (*-NetFirewallRule, WMI) fail during this pass.

UDN

User Defined Network — the OpenShift Virtualization construct used here as a single layer2, non-NAT’d network spanning every Symphony host, management and compute, either site. The network-model invariant the whole design is built around (ADR 0014).

vemkd

Virtual Engine Manager Kernel Daemon — owns EGO_KD_PORT (7870), the first port a Symphony client connects to.

VirtualMachinePool

The KubeVirt CRD managing the Windows compute-host fleet as a scalable pool of identical clones from the golden DataSource (d1 instancetype, dedicated CPU). Also the resource the KEDA demo scaler targets via its standard /scale subresource.

VNI

Virtual Network Interface — the ROKS/IBM Cloud VPC construct (via OVN localnet UDNs) giving a VM a real, VPC-routable IP that bypasses the pod overlay. The mechanism the designed ROKS burst site depends on for NAT-free reachability to the AWS-side masters.

WMCO

Windows Machine Config Operator — the OpenShift feature for running Windows workloads as containers on shared-kernel Windows nodes. Explicitly ruled out for Symphony compute: LIM installs as a persistent Windows service (SCM), not a container entrypoint, and WMCO disallows the privileged, hostile-multi-tenant model Symphony’s grid needs (ADR 0001).