ClusterOS Dashboard

Dashboard

Real-time cluster monitoring and control

Dynamic Tuning

Adjust coefficients in real-time to change graph behavior dynamically

Graph Response

w_u (Utilization Weight) 1.5

k_t (Throughput Boost) 25

k_q (Queue Impact) 20

Payload Intensity

α (Size Weight) 5

β (Sum Divisor) 50

γ (Max Divisor) 20

Time Windows (ms)

Spike Window 2000

Max Job Age 10000

Architecture & Key Concepts

Core distributed systems patterns and runtime mechanisms powering ClusterOS

DNS Routing

Service Discovery and Single System Image
The DNS router provides one entry point for clients, maintains a registry of load balancers, and supports transparent routing so the cluster behaves like one logical system.

Load Balancer Kernel

Priority Queue and Dispatch
The kernel receives requests, orders work by priority, and dispatches jobs to healthy workers while coordinating retries, aggregation, and metrics.

Failure Detection

Phi-Suspicion
Heartbeat intervals are analyzed with a phi-accrual detector so the system can adapt to network delay and identify unhealthy workers without relying on rigid timeouts.

Circuit Breaker

Isolates failing workers across three states, Closed, Open, and Half-Open, to prevent local errors from causing cascading system-wide failures.

Scheduling

Client Affinity and Least-Load Routing
The scheduler keeps sticky client-to-worker affinity when possible, then falls back to healthy least-loaded nodes when the preferred worker is unavailable.

Lamport Clocks

Logical Time Ordering
Logical timestamps preserve causal ordering across distributed messages without requiring synchronized wall-clock time between components.

Job Aggregation

MapReduce
Employs a divide-and-conquer strategy where tasks are partitioned for parallel processing and then reassembled to provide a consistent, ordered result to the client.

Observability

Metrics and Runtime Control
The dashboard and metrics server expose live cluster health, queue depth, throughput, circuit breaker states, and host resource usage for real-time monitoring.

System Controls

Load Balancer

Workers

Cluster Metrics

Healthy

workers

Total

workers

Active

jobs

Queued

jobs

System Health

Cluster Utilization %

Request Throughput (jobs/sec)

Queue Depth (pending jobs)

Status: Waiting for Load Balancer...

Health: --

CPU Usage:

Memory Usage:

Disk Usage:

System Processes: --

Network: -- B↓ / -- B↑

Utilization: --

Worker Health (Circuit Breaker States)

--:--:--

States: ● CLOSED (healthy) ● HALF_OPEN ● OPEN (failed)

No workers connected

Submit Job

Job Payload (JSON Array)

Example: [10, 20, 30] → Elements doubled and distributed across workers

Results

Ready to submit jobs...