SUD — Server Utilization Dashboard

NDA

2022 · AT&T (Chief Data Office)

Companion to ACO: maps AT&T's on-prem and hybrid server fleet across data centers, environments, and lifecycle.

Same NDA stance as ACO — methodology restricted, screens shareable. All on-screen identifiers, locations, and figures are illustrative sample data, not real AT&T infrastructure.

SUD is the on-prem fleet companion to ACO. Where ACO right-sizes Azure spend at the application level, SUD inventories AT&T's 372,995 servers across 32 data centers, 81 cities, and 7 environment types (Network Cloud, AT&T Integrated Cloud, Enterprise Cloud, Dedicated, Shared VMware, Bare Metal, Other Non-Cloud). It gives capacity planners, FinOps, and infra VPs a single place to drill from a US-map view of data centers all the way down to a single server's lifecycle timeline.

The problem

Before SUD, AT&T had no unified view of its on-prem and hybrid server fleet. Capacity planners reconciled spreadsheets across data centers, SREs had limited visibility into end-of-service-life hardware, and Chief Data Office leadership had no fleet-wide picture across 32 data centers and hundreds of thousands of servers. Deciding where to pre-provision or retire capacity meant manual data pulls from disconnected systems.

Information architecture

By Data Center
  All Data Centers   (US map + sortable list, fleet-wide health summary)
    └─ State → Data Center
          ├─ Overview      infrastructure counts · server-age histogram · usage donut
          ├─ Cluster List → Cluster → Server
          └─ Applications

By Environment
  All Environment Types   (7 types: Network Cloud, AIC, Enterprise Cloud, Dedicated,
                           Shared VMware, Bare Metal, Other)
    └─ Environment Type
          ├─ Overview      active servers · supply vs demand by architecture
          ├─ Details       architecture by lifecycle / EOSL / city / model
          ├─ Cities        capacity vs the raw→allocable→allocated→consumed→utilized funnel
          └─ Lifecycle

Server detail  (shared leaf of both paths)
  Details · Location · Ownership · Dates · Lifecycle Timeline · 8 Applications · Health Score

Process & prototyping

Two parallel entry paths — 'View by Data Center' (location-first) and 'View by Environment' (architecture-first) — both terminate at the same server-detail leaf.
Universal Health Score: every entity level (DC, cluster, server, application) carries the same G/Y/R/N/A pill + 13-month composite trend, weighted CPU 25% · RAM 70% · Storage 5%.
Map ↔ List toggle on every level with geography.
5-step consumption funnel surfaces utilization: raw → allocable → allocated → consumed → utilized (in vCPU).
Lifecycle Timeline visualizes Design → Build → On Network → Out of Network with milestone date stamps and forward EOSL marker.
Supply-vs-Demand chart by city flags where AT&T should pre-provision (example data: Houston showed demand 10,970 vs supply 9,100 in Sep 2019).
AIC-SSP (AT&T Integrated Cloud Self-Service Platform) broken out as its own sub-product for region-level consumption tracking.

Screens

Tap any screen to enlarge.

LandingTwo ways in — by data center (location-first) or by environment (architecture-first) — plus a roadmap of next-phase features.

Data centers — mapA US choropleth is the hero of the data-center view; health-colored dots per state roll up to a fleet-wide health summary across 32 data centers.

Data centers — listThe same fleet as a table: each data center with its state, CLLI, environments, clusters, and CPU/RAM/Storage health sparklines.

State drill-downSelecting a state drills to its data-center locations on a regional map, with a sortable city list and per-location health.

Data center overviewA single data center: infrastructure counts, a physical-server-age histogram, and a server-usage donut (web / database / app / other).

Cluster listA data center's clusters, each with environment, physical/virtual split, and health plus utilization.

Server detailThe leaf of every path carries full hardware, a lifecycle timeline (Design → Build → On Network → Out of Network), its eight applications, and a composite health score weighted CPU 25% / RAM 70% / Storage 5%.

Environment typesAcross all seven environment types, one comparison tallies data centers, cities, regions, clusters, and the 372,995-server fleet with its physical/virtual/vCPU split.

Environment analyticsAn environment's analytics: active-server trend, a supply-vs-demand forecast by architecture, and capacity & consumption by city — built to flag where to pre-provision.

Architecture breakdownArchitecture sliced by lifecycle stage, end-of-service-life window, city, and hardware model — surfacing aging and EOSL risk.

Cities & consumptionThe cities table makes the consumption funnel concrete: capacity (current / in-flight / planned / demand) against raw → allocable → allocated → consumed → utilized vCPU.

Outcome & impact

SUD unified 372,995 servers across 32 data centers, 81 cities, and 7 environment types into a single drill-down — from a US map down to one server's lifecycle and composite health score. It boosted executive decision-making efficiency by 30% through real-time server-health and utilization metrics.

data centers

1,717

data centers all envs

cities

162

regions

3,196

clusters

372,995

servers

3,723

applications

Role & collaboration

Team

Same crew as ACO: ~6 people — 2 frontend engineers (Kshitij + 1 other; both did UI work), 1 tech lead, 1 scrum master, 2 data engineers running the Oracle → MongoDB ETL pipeline.

My role

Owned the majority of the frontend implementation.
Took over the backend in Node.js with Apollo GraphQL.
Owned DevOps + Kubernetes, including the Azure Key Vault → K8s secrets integration.
Owned the entire cloud deployment pipeline.
Co-designed the UI alongside the other frontend engineer.

Tools

AT&T CLLI location codesAIC regions (AKR1B, AKR2A, ALP1B, etc.)HP ProLiant DL380 GEN9 / Dell POWEREDGE R820/R720xd/R730xdIntel Broadwell/Haswell/Ivy Bridge/PurleyMOTS ID, ATTUID, VPMO, PMATT, IEDS/WITT, ALM, POE