scenic-srs
by
umang suthar
umang sutharhere is my SRS document.
Landing
Comments (0)
Sign in to leave a comment
Infrastructure OverviewLive · Updated just now
Online
4units→Stable
Total GPUs
4x NVIDIA B300 — On-premise cluster
Degraded
3/ 4↓-1 node
Active Inference Nodes
Node 4 under maintenance — ETA 14 min
In Sync
Sync↑Last: 4m ago
Cloud Backup Status
Redundant offsite backup — 99.9% SLA
Nominal
99.94%↑+0.02%
Uptime
30-day rolling average — 52 min downtime
Live
1,247active↑+8.3%
Current Users
Peak today: 1,412 users at 14:22 UTC
Elevated
284ms↓+41ms
Avg Response Latency
Target: < 250ms — P95: 612ms
GPU Cluster — Cloud Wash On-Premise
4x NVIDIA B300 · NVLink Interconnect · 768 GB Aggregate VRAM
3 / 4 Nominal
GPU-0
NVIDIA B300
Active
VRAM Usage130.6 GB / 192 GB
Temperature61°C
73util%
Power420W / 700W
Model Load
Llama-3.1-70B-Instruct
GPU-1
NVIDIA B300
Busy
VRAM Usage161.3 GB / 192 GB
Temperature74°C
91util%
Power580W / 700W
Model Load
Llama-3.1-70B-Instruct
GPU-2
NVIDIA B300
Active
VRAM Usage99.8 GB / 192 GB
Temperature58°C
55util%
Power310W / 700W
Model Load
DeepSeek-R1-32B
GPU-3
NVIDIA B300
Thermal
VRAM Usage174.7 GB / 192 GB
Temperature88°C
97util%
Power665W / 700W
Model Load
patent-finetune-v3.2
Mac Mini Nodes
API orchestration, queue management, and vector retrieval
3 / 3 Online
mac-mini-01
API Gateway / Load Balancer
CPU38%
RAM14.2 / 24 GB
Uptime42d 07h
mac-mini-02
Inference Queue Manager
CPU62%
RAM19.8 / 24 GB
Uptime42d 07h
mac-mini-03
Vector DB / RAG Service
CPU44%
RAM16.4 / 24 GB
Uptime41d 18h
Failover Status
On-Premise → Cloud backup readiness
Failover Ready
Primary Inference
ActiveLive uptime
32d 23h 00m 14sAvailability SLA
99.97%Last health check
12s agoModel
Llama-3.1-70B-LoRAPrimary inference engine active00:12 ago
GPU cluster health verified (4x B300)00:12 ago
Internal network latency < 2ms01:00 ago
NVMe model storage mounted05:30 ago
Standby Replica
StandbyReplica model status
ReadyBackup verified at
2026-05-05 04:00 UTCReplication lag
~4.2sEst. failover time
< 90sReplica model weights synced03:22 ago
Backup snapshot verified (S3)15:00 ago
DNS failover route pre-warmed08:10 ago
TLS certificate valid01:00:00 ago
Auto-Failover
Will activate if on-prem goes offline
Failover Simulation
Run a non-destructive drill — routes a shadow request to the cloud replica and verifies RTO < 90s without interrupting live traffic.
Backup & Redundancy
All Systems Protected
RPO
1 hr
Recovery Point Objective
RTO
4 hrs
Recovery Time Objective
Last Backup
May 5, 02:01 AM
18 min · 147 GB transferred
Total Backup Size
1.02 TB
Across 3 redundancy locations
Last 7 Backups
Success
Failed
Redundancy Locations
On-Premise NAS
Cloud Wash GPU Facility
AWS S3 (us-east-1)
N. Virginia Region
Backblaze B2
Sacramento, CA
Current Backup Size
147GB
Model Weights98.4 GB
Documents31.2 GB
Snapshots17.4 GB
Storage Allocation
Model Weights & Checkpoints68%
Patent Document Archives42%
System Snapshots27%
System AlertsLive
2 Critical3 Warnings2 Info
critical
critical
warning
warning
warning
info
info
System Health Overview
Real-time infrastructure diagnostics — GPU inference cluster & services
Updated 12s ago
67%
67%
CPU Load
Avg across 4x B300 GPUs
43%
43%
Memory Usage
348 GB / 800 GB used
Disk Space
15.6 TB / 20 TB used
Queue Depth
44 / 200 jobs queued
Network I/O Rate
Upload
1.24 Gbps
Download
0.87 Gbps
Healthy (0–60%)Caution (61–85%)Critical (>85%)
Scaling & Capacity Planning
GPU cluster growth from 4 → 35+ nodes
Auto-ScalingDisabled
Current Cluster Capacity
4
GPUs Active
Target: 35+ GPUs by Q2 2027
GPU-01 — 94% util
GPU-02 — 87% util
GPU-03 — 91% util
GPU-04 — 76% util
GPU-05 — Pending
GPU-06 — Pending
GPU-07 — Pending
Future node
Future node
Future node
Future node
Cost Projections
Current / Mo
$6,200
4 GPU nodes
Phase 2 / Mo
$10,850
7 GPU nodes
Phase 3 / Mo
$21,700
14 GPU nodes
Peak / Mo
$54,250
35 GPU nodes
Cloud Wash GPU rental includes heat reuse credit — effective cost offset of ~12% vs bare-metal cloud.
GPU Demand Forecast — 2026
Actual
Projected
Planned Cap.
Scaling Timeline — 4 GPUs to 35+ GPUs
Est. completion: Q2 2027
Phase 1 — Baseline (Current)
4x NVIDIA B300 GPUs active. Serving ~120 concurrent inference sessions at 94% avg utilization.
Phase 2 — Initial Scale-Out
Provision 3 additional B300 nodes. Target: 7 GPUs, supporting 210 concurrent sessions.
Phase 3 — Mid Growth
Horizontal pod autoscaling enabled. Scale to 14 GPUs across 2 availability zones.
Phase 4 — Peak Capacity
Distributed inference cluster at 35 GPUs. Multi-site load balancing across Cloud Wash facilities.
Scaling Recommendations
GPU Utilization Above 90% Threshold
Average GPU load at 87% — approaching saturation. Recommend provisioning Phase 2 nodes within 14 days.
Enable Kubernetes HPA for Inference Pods
Horizontal Pod Autoscaler not configured for inference workloads. Enables burst capacity during peak patent drafting demand.
Reserved Instance Pricing Available
Committing to 12-month reserved GPU instances reduces projected cloud backup costs by ~34% ($2,100/mo savings).
System Logs
Inference server, GPU manager, RAG pipeline, and service logs
Live
| Timestamp | Level | Service | Message | Log ID |
|---|---|---|---|---|
| 2026-05-05 14:32:07 | ERROR | GPU Manager | LOG-00421 | |
| 2026-05-05 14:31:54 | WARNING | API Server | LOG-00420 | |
| 2026-05-05 14:31:47 | INFO | RAG Pipeline | LOG-00419 | |
| 2026-05-05 14:31:33 | INFO | Auth Service | LOG-00418 | |
| 2026-05-05 14:31:20 | DEBUG | GPU Manager | LOG-00417 | |
| 2026-05-05 14:31:08 | ERROR | USPTO Bridge | LOG-00416 | |
| 2026-05-05 14:30:55 | INFO | API Server | LOG-00415 | |
| 2026-05-05 14:30:41 | WARNING | RAG Pipeline | LOG-00414 | |
| 2026-05-05 14:30:29 | DEBUG | Scheduler | LOG-00413 | |
| 2026-05-05 14:30:17 | INFO | GPU Manager | LOG-00412 |
No comments yet. Be the first!