Cloud Compute & Sandboxes Architecture: The Complete Guide

High-performance GPU instances, elastic CPU clusters, and sub-second Firecracker microVM sandboxes engineered for AI model inference, agent execution, and large-scale data processing.

FOTOhub Compute bridges direct AWS EC2 cloud infrastructure managed by the Compute Engine (server/compute-engine/ on port 8801) with lightweight multi-tenant virtualization managed by the Agent Compute Engine (server/agent-compute/ on port 8795).

1. Introduction: What is FOTOhub Compute and Who is it For?

FOTOhub Compute is a comprehensive, API-first infrastructure platform designed specifically for modern AI builders, data scientists, and backend engineers. It provides a seamless, programmatically-driven experience for provisioning and managing computational resources, ranging from lightweight CPU microVMs to massive multi-GPU training clusters.

1.1 The FOTOhub Compute Philosophy

We believe that developers should have access to the raw power of AWS without the complexity of IAM roles, VPC configurations, and unpredictable billing. FOTOhub Compute gives you the exact same hardware, the exact same reliability, but with a simplified API, an integrated prepaid wallet, and built-in guardrails to prevent surprise bills.

By abstracting away the tedious aspects of cloud infrastructure, we allow you to focus on what matters most: building incredible AI applications, processing massive datasets, and deploying robust backend services.

1.2 Target Audience and Use Cases

AI/ML Engineers & Researchers

Train, fine-tune, and deploy large language models (LLMs), diffusion models (like Stable Diffusion XL and Flux), and complex computer vision pipelines. Utilize our high-end NVIDIA GPUs (A10G, T4) for unparalleled performance and cost-efficiency.

Autonomous Agent Developers

Run autonomous AI agents in secure, isolated environments with sub-200ms cold starts using our Firecracker microVM sandboxes. Perfect for executing untrusted AI-generated code, running evaluation loops, and handling asynchronous tool calls securely.

Data Engineers & Big Data Practitioners

Process large datasets, transcode videos, and run massive batch jobs using our scalable CPU clusters. Leverage our Compute (C5), Memory (R5), and General Purpose (M5, T3) families to optimize for your specific data processing bottlenecks.

Backend & Full-Stack Developers

Host reliable web services, APIs, microservices, and databases on cost-effective infrastructure. Enjoy built-in storage integrations, networking capabilities, and one-click DNS management to simplify your deployment lifecycle.

1.3 Key Platform Capabilities

Diverse Hardware Selection: Choose from 22 distinct instance types across 6 families (T3, C5, M5, R5, G4dn, G5), ensuring the perfect match for any workload footprint.
Transparent, Predictable Billing: 100% USD prepaid wallet system. No surprise invoices, no complex credit conversions. You control your exact maximum spend.
Spot Instance Arbitrage: Leverage unused AWS cloud capacity at massive discounts (up to 70% off on-demand prices) with automated fallback mechanisms. FOTOhub passes spot market prices directly through to you without markups.
Instant Provisioning & Scalability: Launch fully-configured GPU instances in minutes and Firecracker sandboxes in milliseconds.
Secure by Default: Every instance is deployed in an isolated security group, featuring KMS-encrypted storage, ephemeral SSH keys, and continuous billing enforcement to protect your wallet and your data.
Seamless Storage Integration: Attach high-performance EBS volumes (gp3, io2) or stream data directly to/from FOTOhub S3 (at $0.00 intra-cluster egress) or your own external BYOB buckets (AWS, Cloudflare R2, GCP).

2. High-Level Architecture & Control Plane

The FOTOhub Compute architecture is meticulously designed for global scale, strict security, and ultra-low latency. It consists of multiple interoperating layers, spanning from the developer-facing API down to the physical hardware in our Frankfurt (eu-central-1) data center.

mermaid

flowchart TD
    subgraph Client Layer ["Client & API Access Layer"]
        direction TB
        A["Developer API / REST Endpoints"]
        B["Console UI (fotohub.app/console)"]
        C["AI Agent System (Claude Opus / DeepSeek)"]
        CLI["FOTOhub CLI Tooling (Python, TS, Go)"]
        SDK["Official SDKs"]
    end

    subgraph Compute Control Plane ["FOTOhub Control Plane (Orchestration & Billing)"]
        direction TB
        D["Compute Engine API (Port 8801) - EC2 Management"]
        E["Agent Compute Orchestrator (Port 8795) - microVMs"]
        F["Prepaid USD Wallet Billing Engine (Real-time)"]
        DNS["Route53 DNS Manager & Domain Engine"]
        SEC["Security, IAM & Firewall Policy Manager"]
        MTR["Telemetry & Metering Service"]
    end

    subgraph Dedicated Cloud Infrastructure ["AWS eu-central-1 (Frankfurt) Physical Layer"]
        direction TB
        G["EC2 G-Series GPU Fleet (A10G, T4 - Spot & On-Demand)"]
        H["EC2 CPU Fleet (T3, C5, M5, R5)"]
        I["EBS High-Speed Persistent Storage (gp3 / io2)"]
        J["Elastic IPs & Security Group Firewalls"]
        K["CloudWatch Telemetry & Serial Console Output"]
        KMS["AWS KMS Encryption Engine (XTS-AES-256)"]
        VPC["Dedicated Multi-Tenant VPC Subnets"]
    end

    subgraph Lightweight Virtualization Plane ["Agent Virtualization Plane"]
        direction TB
        L["Firecracker microVM Pool (<200ms cold start)"]
        M["vsock Host-Guest Communication Channel (Port 9999)"]
        N["Virtual Workspace Mounts (/data/workspaces/{user_id})"]
        O["Real-Time SSE Event Bus (agent_delta, commentary)"]
        P["Jailbreak Detection & Anomaly Monitoring"]
        SANDBOX["Ephemeral Sandbox Execution Contexts"]
    end

    %% Connections
    A & B & CLI & SDK --> D
    C --> E
    D --> F
    E --> F
    D --> MTR
    E --> MTR
    
    D --> G & H & I & J & K & DNS & KMS & VPC
    E --> L & M & N & O & P & SANDBOX
    
    G & H <--> I
    L <--> N
    
    classDef client fill:#f9f,stroke:#333,stroke-width:2px;
    classDef control fill:#bbf,stroke:#333,stroke-width:2px;
    classDef infra fill:#bfb,stroke:#333,stroke-width:2px;
    classDef virt fill:#fbb,stroke:#333,stroke-width:2px;
    
    class A,B,C,CLI,SDK client;
    class D,E,F,DNS,SEC,MTR control;
    class G,H,I,J,K,KMS,VPC infra;
    class L,M,N,O,P,SANDBOX virt;

2.1 The Compute Engine (EC2 Abstraction)

Operating on Port 8801, the Compute Engine handles the heavy lifting of EC2 provisioning. It interfaces directly with AWS APIs to launch instances, attach EBS volumes, allocate Elastic IPs, and configure Route53 DNS records.

2.2 The Agent Compute Orchestrator (MicroVMs)

Operating on Port 8795, this orchestrator manages our pool of Firecracker microVMs. It maintains warm pools of stripped-down Linux kernels to guarantee sub-200ms cold starts for ephemeral task execution, making it perfect for rapid AI agent tool calls.

3. Comprehensive Hardware Specifications & 22-Instance Catalog

All compute instances physically reside in Frankfurt (eu-central-1) across multiple Availability Zones (eu-central-1a, eu-central-1b, eu-central-1c). This strategic location ensures ultra-low latency connections to European storage, major internet exchanges, and global inference backbones.

Dynamic Pricing Notice

Spot prices fluctuate based on real-time AWS market demand. FOTOhub passes spot market prices directly through to your prepaid USD wallet without any markups. The prices shown below are estimates based on typical market conditions. For exact, real-time pricing, always query the Live Catalog API.

3.1 GPU Accelerated Instances (NVIDIA)

Optimized for generative AI, PyTorch model training, vLLM inference, ComfyUI pipelines, 3D rendering, and massively parallel workloads.

Catalog ID	GPU Model	GPUs	VRAM	vCPU	RAM	Bandwidth	On-Demand ($/hr)	Spot ($/hr)*	Best Use Case / Savings
`g4dn.xlarge`	NVIDIA T4	1	16 GB	4	16 GB	Up to 25 Gbps	$0.53	$0.20	Lightweight Inference, SD1.5, Audio Processing (62% savings)
`g4dn.2xlarge`	NVIDIA T4	1	16 GB	8	32 GB	Up to 25 Gbps	$0.75	$0.27	Mid-tier Inference, Data Prep, Whisper Speech-to-Text (64% savings)
`g4dn.4xlarge`	NVIDIA T4	1	16 GB	16	64 GB	Up to 25 Gbps	$1.20	$0.48	Heavy CPU Pre-processing + GPU Inference (60% savings)
`g4dn.8xlarge`	NVIDIA T4	1	16 GB	32	128 GB	50 Gbps	$2.18	$0.85	Massive Data Pipelines with GPU acceleration (61% savings)
`g4dn.12xlarge`	NVIDIA T4	4	64 GB	48	192 GB	50 Gbps	$3.91	$1.52	Multi-GPU Distributed Inference (vLLM) (61% savings)
`g4dn.16xlarge`	NVIDIA T4	1	16 GB	64	256 GB	50 Gbps	$4.35	$1.70	Extreme CPU + Light GPU workloads (60% savings)
`g5.xlarge`	NVIDIA A10G	1	24 GB	4	16 GB	Up to 25 Gbps	$1.01	$0.38	SDXL, ComfyUI, 7B-14B LLM Inference (62% savings)
`g5.2xlarge`	NVIDIA A10G	1	24 GB	8	32 GB	Up to 25 Gbps	$1.20	$0.45	LoRA Training, 14B LLM Fine-tuning (63% savings)
`g5.4xlarge`	NVIDIA A10G	1	24 GB	16	64 GB	Up to 25 Gbps	$1.61	$0.60	3D Rendering, Heavy Video Transcoding (63% savings)
`g5.8xlarge`	NVIDIA A10G	1	24 GB	32	128 GB	Up to 25 Gbps	$2.28	$0.85	Massive Batch Inference, Data Augmentation (62% savings)
`g5.12xlarge`	NVIDIA A10G	4	96 GB	48	192 GB	50 Gbps	$5.67	$2.15	Multi-GPU LLM Training (30B+ params) (62% savings)
`g5.24xlarge`	NVIDIA A10G	4	96 GB	96	384 GB	100 Gbps	$8.14	$3.10	High-Throughput Enterprise Inference (61% savings)
`g5.48xlarge`	NVIDIA A10G	8	192 GB	192	768 GB	100 Gbps	$16.29	$6.20	Foundation Model Training (61% savings)

3.2 General Purpose CPU Instances (T3 & M5 Families)

Ideal for backend services, web servers, microservices, development environments, and applications with balanced compute, memory, and networking needs.

Catalog ID	Family Focus	vCPU	RAM	Bandwidth	On-Demand ($/hr)	Spot ($/hr)	Best Use Case
`t3.micro`	Burstable	2	1 GB	Up to 5 Gbps	$0.010	$0.003	Bastion Hosts, Cron Job Runners
`t3.small`	Burstable	2	2 GB	Up to 5 Gbps	$0.021	$0.006	Light Web Servers, API Gateways
`t3.medium`	Burstable	2	4 GB	Up to 5 Gbps	$0.042	$0.013	Dev Environments, Discord Bots
`t3.large`	Burstable	2	8 GB	Up to 5 Gbps	$0.084	$0.025	CI/CD Runners, Testing Servers
`t3.xlarge`	Burstable	4	16 GB	Up to 5 Gbps	$0.168	$0.050	Mid-size CI/CD, General purpose workers
`t3.2xlarge`	Burstable	8	32 GB	Up to 5 Gbps	$0.336	$0.114	Large compile jobs, heavy build servers
`m5.large`	Consistent	2	8 GB	Up to 10 Gbps	$0.096	$0.037	Production Microservices, Small web nodes
`m5.xlarge`	Consistent	4	16 GB	Up to 10 Gbps	$0.192	$0.075	Small Databases, Message Queues (RabbitMQ)
`m5.2xlarge`	Consistent	8	32 GB	Up to 10 Gbps	$0.384	$0.150	High-Traffic Web Applications, Kafka nodes
`m5.4xlarge`	Consistent	16	64 GB	Up to 10 Gbps	$0.768	$0.300	Distributed Data Processing, Cache clusters

3.3 Compute & Memory Optimized Instances (C5 & R5 Families)

Designed for specialized workloads requiring high CPU throughput (C5) or massive memory footprints for in-memory processing (R5).

Catalog ID	Family Focus	vCPU	RAM	Bandwidth	On-Demand ($/hr)	Spot ($/hr)	Best Use Case
`c5.large`	Compute	2	4 GB	Up to 10 Gbps	$0.085	$0.032	Batch Processing, Log Parsing, Web Scrapers
`c5.xlarge`	Compute	4	8 GB	Up to 10 Gbps	$0.170	$0.064	Video Encoding (FFmpeg), CPU Inference
`c5.2xlarge`	Compute	8	16 GB	Up to 10 Gbps	$0.340	$0.128	High-Performance Web Servers, Load Balancers
`c5.4xlarge`	Compute	16	32 GB	Up to 10 Gbps	$0.680	$0.256	Scientific Modeling, Massive Compiling
`r5.large`	Memory	2	16 GB	Up to 10 Gbps	$0.126	$0.047	Redis/Memcached Clusters, Light databases
`r5.xlarge`	Memory	4	32 GB	Up to 10 Gbps	$0.252	$0.095	In-Memory Databases, Search Indices
`r5.2xlarge`	Memory	8	64 GB	Up to 10 Gbps	$0.504	$0.190	Vector Databases (Pinecone/Milvus hosting)
`r5.4xlarge`	Memory	16	128 GB	Up to 10 Gbps	$1.008	$0.380	Real-Time Big Data Analytics, Spark Clusters

4. Querying the Live Catalog API

To retrieve real-time pricing, exact hardware specifications, and availability across zones, always use the Live Catalog API.

http

GET /compute/v1/catalog
Authorization: Bearer fh_live_YOUR_API_KEY

Response Example

json

{
  "catalog": [
    {
      "id": "g5.xlarge",
      "family": "g5",
      "gpu": {
        "model": "NVIDIA A10G",
        "count": 1,
        "vram_gb": 24,
        "cuda_cores": 9216,
        "tensor_cores": 288
      },
      "cpu": {
        "cores": 4,
        "arch": "x86_64",
        "model": "AMD EPYC 7R32"
      },
      "memory_gb": 16,
      "network_bandwidth_gbps": 25,
      "pricing": {
        "currency": "USD",
        "on_demand_hourly": 1.012,
        "spot_hourly": 0.383,
        "spot_savings_percent": 62.1
      },
      "available_zones": ["eu-central-1a", "eu-central-1b"]
    }
  ]
}

PythonTypeScriptcURL

python

from fotohub import FotoHub

client = FotoHub()
catalog = client.get("/compute/v1/catalog")

print("Available GPU Instances:")
for item in catalog["catalog"]:
    if item.get("gpu"):
        print(f"[{item['id']}] {item['gpu']['model']} ({item['gpu']['vram_gb']}GB VRAM)")
        print(f"  Spot Rate: ${item['pricing']['spot_hourly']}/hr")
        print(f"  Savings vs On-Demand: {item['pricing']['spot_savings_percent']}%\n")

typescript

import { FotoHub } from "fotohub";

const client = new FotoHub();
const { data } = await client.get("/compute/v1/catalog");

console.log("Available GPU Instances:");
data.catalog.forEach((item: any) => {
  if (item.gpu) {
    console.log(`[${item.id}] ${item.gpu.model} (${item.gpu.vram_gb}GB VRAM)`);
    console.log(`  Spot Rate: $${item.pricing.spot_hourly}/hr`);
    console.log(`  Savings vs On-Demand: ${item.pricing.spot_savings_percent}%\n`);
  }
});

bash

curl -X GET https://apis.fotohub.app/compute/v1/catalog   -H "Authorization: Bearer $FOTOHUB_API_KEY"   | jq '.catalog[] | select(.gpu != null) | {id, model: .gpu.model, spot: .pricing.spot_hourly}'

5. Infrastructure Comparison: FOTOhub vs Alternatives

Why choose FOTOhub Compute over direct AWS access or specialized AI clouds like RunPod and Lambda Labs? We provide the perfect balance of enterprise-grade reliability, transparent pricing, and developer experience.

Feature / Metric	FOTOhub Compute	AWS Direct	RunPod / Vast	Lambda Labs
A10G Spot Rate	$0.38 / hr	~$0.38 / hr	N/A (Server-grade)	N/A
AWS Cloud Resilience	Yes (Tier-1 Datacenter)	Yes	Variable (Crowdsourced)	Yes (Tier-2)
Sub-200ms Firecracker microVMs	Built-in	AWS Lambda only	No	No
Root SSH Key Injection	Automated	Manual Key Pair	Custom SSH keys	Manual Key Pair
Route53 DNS Integration	One-click API	Complex IAM Setup	No	No
Hot EBS Volume Resize	Online (`PUT /volumes`)	AWS CLI / Console	No (Requires restart)	No
Unified USD Prepaid Wallet	Yes (1:1 pass-through)	Enterprise Invoicing	Credit Card	Card / Credits
Automatic Safety Shutdown	Hard `max_runtime_hours`	Manual CloudWatch Alarms	Manual	Manual
Provisioning Speed	< 60 seconds	Minutes	Seconds to Minutes	Minutes (if available)
Minimum Spend Requirements	$0.50 per instance	None (but infinite risk)	Variable	Hourly minimums
Security Groups / Firewalls	API-Driven, Strict Default	Complex VPC Setup	Basic Port Mapping	Basic Port Mapping

5.1 The FOTOhub Advantage

FOTOhub acts as an intelligent abstraction layer over AWS infrastructure. You get the raw power, reliability, and security of AWS without the DevOps overhead of managing VPCs, IAM roles, and complex billing alarms. Our transparent pass-through pricing on Spot instances means you pay exactly what AWS charges, funded directly from your prepaid wallet. No markup, no hidden fees.

6. Quick Start in 5 Minutes

Launch a GPU instance, retrieve your ephemeral SSH key, and connect to it in under 5 minutes.

6.1 Prerequisites

A FOTOhub Account with a positive USD Wallet balance (Minimum $0.50 USD required to provision).
An API Key (starts with fh_live_).

6.2 Step 1: Provision an Instance

We will launch a g5.xlarge instance using Spot pricing to maximize savings. We'll set an auto-stop hour to prevent runaway costs.

PythonTypeScriptGocURL

python

import os
import requests
import time
import sys

API_KEY = os.environ.get("FOTOHUB_API_KEY")
if not API_KEY:
    print("Please set FOTOHUB_API_KEY environment variable")
    sys.exit(1)

HEADERS = {"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"}
BASE_URL = "https://apis.fotohub.app/compute/v1"

# 1. Request a new instance
payload = {
    "instance_type": "g5.xlarge",
    "market_type": "spot",
    "disk_size_gb": 100,
    "image_id": "ubuntu-22.04-gpu-optimized",
    "auto_stop_hours": 4  # Prevent runaway costs! Automatically stops after 4 hours.
}

print("Initiating provisioning request...")
response = requests.post(f"{BASE_URL}/instances", json=payload, headers=HEADERS)
response.raise_for_status()

instance = response.json()
instance_id = instance["id"]

print(f"Successfully requested instance. ID: {instance_id}")
print("Polling for running state...")

# 2. Poll until running
while True:
    status_resp = requests.get(f"{BASE_URL}/instances/{instance_id}", headers=HEADERS).json()
    state = status_resp["state"]
    print(f"Current Status: {state}")
    
    if state == "running":
        ip_address = status_resp["public_ip"]
        print(f"\n✅ Instance is fully running!")
        print(f"Public IP Address: {ip_address}")
        break
    elif state in ["failed", "terminated"]:
        print("\n❌ Provisioning failed or instance was terminated early.")
        sys.exit(1)
        
    time.sleep(5)

typescript

import axios from 'axios';

const API_KEY = process.env.FOTOHUB_API_KEY;
if (!API_KEY) {
  console.error("Please set FOTOHUB_API_KEY environment variable");
  process.exit(1);
}

const HEADERS = { Authorization: `Bearer ${API_KEY}` };
const BASE_URL = 'https://apis.fotohub.app/compute/v1';

async function launchInstance() {
  try {
    console.log("Initiating provisioning request...");
    // 1. Request instance
    const { data: instance } = await axios.post(`${BASE_URL}/instances`, {
      instance_type: 'g5.xlarge',
      market_type: 'spot',
      disk_size_gb: 100,
      image_id: 'ubuntu-22.04-gpu-optimized',
      auto_stop_hours: 4
    }, { headers: HEADERS });
    
    console.log(`Successfully requested instance. ID: ${instance.id}`);
    console.log("Polling for running state...");
    
    // 2. Poll status
    let state = 'pending';
    let ipAddress = '';
    
    while (state !== 'running') {
      await new Promise(resolve => setTimeout(resolve, 5000));
      const { data: status } = await axios.get(`${BASE_URL}/instances/${instance.id}`, { headers: HEADERS });
      state = status.state;
      console.log(`Current Status: ${state}`);
      
      if (state === 'failed' || state === 'terminated') {
        throw new Error('Provisioning failed or instance was terminated early.');
      }
      ipAddress = status.public_ip;
    }
    
    console.log(`\n✅ Instance is fully running!`);
    console.log(`Public IP Address: ${ipAddress}`);
  } catch (error) {
    console.error("Error launching instance:", error.message);
  }
}

launchInstance();

package main

import (
	"bytes"
	"encoding/json"
	"fmt"
	"io/ioutil"
	"net/http"
	"os"
	"time"
)

func main() {
	apiKey := os.Getenv("FOTOHUB_API_KEY")
	if apiKey == "" {
		fmt.Println("Please set FOTOHUB_API_KEY environment variable")
		os.Exit(1)
	}
	baseURL := "https://apis.fotohub.app/compute/v1"

	payload := map[string]interface{}{
		"instance_type":   "g5.xlarge",
		"market_type":     "spot",
		"disk_size_gb":    100,
		"image_id":        "ubuntu-22.04-gpu-optimized",
		"auto_stop_hours": 4,
	}
	jsonPayload, _ := json.Marshal(payload)

	fmt.Println("Initiating provisioning request...")
	req, _ := http.NewRequest("POST", baseURL+"/instances", bytes.NewBuffer(jsonPayload))
	req.Header.Set("Authorization", "Bearer "+apiKey)
	req.Header.Set("Content-Type", "application/json")

	client := &http.Client{}
	resp, err := client.Do(req)
	if err != nil { panic(err) }
	defer resp.Body.Close()

	body, _ := ioutil.ReadAll(resp.Body)
	var result map[string]interface{}
	json.Unmarshal(body, &result)

	instanceID := result["id"].(string)
	fmt.Printf("Successfully requested instance. ID: %s\n", instanceID)
	fmt.Println("Polling for running state...")

	for {
		req, _ = http.NewRequest("GET", baseURL+"/instances/"+instanceID, nil)
		req.Header.Set("Authorization", "Bearer "+apiKey)
		resp, _ = client.Do(req)
		
		body, _ = ioutil.ReadAll(resp.Body)
		resp.Body.Close()
		
		var status map[string]interface{}
		json.Unmarshal(body, &status)
		
		state := status["state"].(string)
		fmt.Printf("Current Status: %s\n", state)
		
		if state == "running" {
			fmt.Println("\n✅ Instance is fully running!")
			fmt.Printf("Public IP Address: %s\n", status["public_ip"].(string))
			break
		} else if state == "failed" || state == "terminated" {
			fmt.Println("\n❌ Provisioning failed or instance was terminated early.")
			os.Exit(1)
		}
		time.Sleep(5 * time.Second)
	}
}

bash

# 1. Provision Instance
curl -s -X POST https://apis.fotohub.app/compute/v1/instances   -H "Authorization: Bearer $FOTOHUB_API_KEY"   -H "Content-Type: application/json"   -d '{
    "instance_type": "g5.xlarge",
    "market_type": "spot",
    "disk_size_gb": 100,
    "image_id": "ubuntu-22.04-gpu-optimized",
    "auto_stop_hours": 4
  }' | jq .

# Note the ID returned, e.g., {"id": "inst_1234567890", "state": "pending", ...}

# 2. Poll Status (Run this until state is "running")
curl -s -X GET https://apis.fotohub.app/compute/v1/instances/inst_1234567890   -H "Authorization: Bearer $FOTOHUB_API_KEY" | jq '{state: .state, public_ip: .public_ip}'

6.3 Step 2: Retrieve Ephemeral SSH Key and Connect

Once the instance reaches the running state, you must retrieve the ephemeral SSH key to access it. FOTOhub does not use static key pairs.

Critical Security Notice

For security reasons, this key can only be downloaded exactly once. If you lose the key or close the terminal before saving it, you will need to terminate the instance and provision a new one. The API will permanently delete the private key from memory immediately after you download it.

cURL

bash

# Download the private key securely and save it
curl -s -X POST https://apis.fotohub.app/compute/v1/instances/inst_1234567890/ssh-key   -H "Authorization: Bearer $FOTOHUB_API_KEY"   | jq -r .private_key > ~/.ssh/fotohub_inst_1234567890.pem

# Set stringent read-only permissions (required by SSH clients)
chmod 400 ~/.ssh/fotohub_inst_1234567890.pem

# Connect to the instance using the 'ubuntu' user
ssh -i ~/.ssh/fotohub_inst_1234567890.pem ubuntu@<PUBLIC_IP_ADDRESS>

7. Billing & USD Wallet Deep Dive

FOTOhub operates entirely on a prepaid USD wallet system. This ensures you never receive an unexpected bill at the end of the month.

Important Note: We do not support credits, promotional tokens, or alternative currencies (NO PLN, NO EUR). All transactions, billing metrics, and api charges are settled exclusively in USD.

7.1 The Billing Lifecycle Flow

mermaid

stateDiagram-v2
    [*] --> CheckBalance: API Provision Request (POST /instances)
    CheckBalance --> Reject: Balance < $0.50 USD
    CheckBalance --> Provision: Balance >= $0.50 USD
    
    Provision --> Running: Instance Booted Successfully
    Running --> Metering: Continuous check (Every 60 seconds)
    
    Metering --> DeductFunds: Calculate Cost (Uptime * Rate)
    DeductFunds --> LowBalanceWarning: Balance drops below $5.00 USD
    DeductFunds --> Running: Balance Healthy (> $5.00 USD)
    
    LowBalanceWarning --> NotifyUser: Send Email / Trigger Webhook
    LowBalanceWarning --> AutoStop: Balance reaches <= $0.00 USD
    
    AutoStop --> Stopped: Instance Gracefully Halted (Data saved to EBS)
    Stopped --> Terminated: 7 days inactive or Manual Delete request
    Terminated --> [*]: All Resources Released

7.2 Key Billing Rules & Enforcements

Minimum Provisioning Balance: You must have at least $0.50 USD in your wallet to provision a new instance or sandbox. API requests will fail with a 402 Payment Required if you are below this threshold.
Per-Second Billing Granularity: Instances are billed per second, with a minimum charge of 60 seconds per boot cycle.
Spot Pricing Pass-Through: If AWS Spot prices fluctuate during your workload, your hourly rate fluctuates accordingly. We do not mark up spot prices. You pay exactly what AWS charges.
Auto-Stop on Zero Balance: If your wallet hits exactly $0.00 USD, running instances are immediately and gracefully halted (ec2.stop_instance).
- Note: Attached EBS volumes remain intact after a stop. They will continue to incur minimal storage costs. If your balance remains negative, volumes will be deleted after 7 days.
EBS Storage Costs: Persistent storage is billed regardless of whether the attached instance is running or stopped.
Network Egress (Bandwidth):
- Inbound traffic: Free ($0.00/GB).
- Outbound to Public Internet: $0.09/GB.
- Outbound to FOTOhub S3 (s1.fotohub.app): Free ($0.00/GB).

Best Practice: Prevent Runaway Costs

Always utilize the auto_stop_hours or max_runtime_hours parameters when provisioning instances via the API. This hard limit prevents instances from running indefinitely if you forget to shut them down, safeguarding your wallet balance.

8. Security Model Deep Dive

Security is a foundational pillar of FOTOhub Compute. We operate under the assumption that instances will be running untrusted AI code, experimental models, or handling sensitive corporate datasets.

8.1 Multi-Layered Perimeter Defense

Isolated Security Groups (VPC Isolation):
- Every user operates within an isolated subnet context. Your instances cannot communicate with other tenants' instances on the local network.
- By default, newly provisioned instances are locked down. Only Port 22 (SSH) is open to the internet.
- To expose a web service (e.g., ComfyUI on 8188, vLLM on 8000, Jupyter on 8888), you must explicitly use the FOTOhub Firewall API (POST /compute/v1/firewall/rules) to open specific ports.
KMS-Encrypted Disk Storage:
- All root EBS volumes and dynamically attached secondary volumes are encrypted at rest using AWS KMS (Key Management Service).
- We utilize the industry standard XTS-AES-256 block cipher encryption.
- Volumes cannot be detached and read by unauthorized AWS accounts or even by FOTOhub support staff.
Automated Ephemeral SSH Keys:
- We completely eliminate the need for static IAM key pairs.
- During instance boot, FOTOhub injects a dynamically generated, KMS-encrypted SSH public key into the guest OS.
- The private key is held in memory by the API and can only be downloaded exactly once by the authenticated user who created the instance.
Metadata Service Protection:
- Instances are strictly shielded from accessing sensitive AWS Instance Metadata Service (IMDSv2) endpoints. This prevents SSRF (Server-Side Request Forgery) attacks from leaking underlying cloud credentials.
Continuous Billing Enforcement:
- Financial security is just as important. Instances are continuously metered against your available USD balance. If your balance is exhausted, instances are gracefully stopped rather than incurring massive runaway debt.

9. Integrated Storage & Data Pipelines

Compute is only half the equation; moving multi-gigabyte models, massive datasets, and high-resolution video quickly is crucial for AI workloads. FOTOhub Compute is tightly integrated with our persistent block and object storage hierarchy.

mermaid

flowchart LR
    subgraph Compute Node
        direction TB
        A["EC2 GPU Instance (eu-central-1)"]
        RAM["System RAM / GPU VRAM"]
    end
    
    subgraph Block Storage
        direction TB
        B["EBS gp3 Persistent Volumes"]
        B2["EBS io2 High-IOPS Volumes"]
    end
    
    subgraph Object Storage
        direction TB
        C["FOTOhub S3 (s1.fotohub.app)"]
        D["BYOB External Destinations"]
        E["AWS S3 / Cloudflare R2 / GCS / Supabase"]
    end
    
    A <-->|NVMe PCI-e (Ultra-Low Latency)| B
    A <-->|NVMe PCI-e (Ultra-Low Latency)| B2
    A <-->|Zero Egress API (High Throughput)| C
    A -->|Direct Stream| D
    D --> E
    
    style A fill:#f96,stroke:#333,stroke-width:2px
    style B fill:#69f,stroke:#333,stroke-width:2px
    style B2 fill:#69f,stroke:#333,stroke-width:2px
    style C fill:#9f6,stroke:#333,stroke-width:2px
    style D fill:#ccc,stroke:#333,stroke-width:2px
    style E fill:#fcc,stroke:#333,stroke-width:2px

9.1 Elastic Block Store (EBS) Persistent Volumes

Attach persistent NVMe disks up to 4 TB for storing persistent model weights (e.g., /data/models), large training datasets, or maintaining state across instance reboots.

gp3 (General Purpose): $0.08/GB-month. Delivers a solid baseline performance of 3,000 IOPS and 125 MB/s throughput, scalable up to 1,000 MB/s. Best for standard ML training, model caching, and typical web workloads.
io2 (Provisioned IOPS): $0.125/GB-month. Designed for mission-critical, latency-sensitive databases and high-throughput streaming. Scales up to a massive 64,000 IOPS and 1,000 MB/s throughput.

Hot Volume Resizing

You can dynamically increase the size of an attached volume without stopping your instance or unmounting the drive. Simply use the PUT /compute/v1/volumes/{id} API to expand capacity on the fly.

9.2 FOTOhub S3 Cloud Storage

Our native, fully S3-compatible managed object storage located in the Frankfurt region.

Pricing: Flat $0.0245/GB-month.
Bandwidth: $0.00 intra-cluster egress. Transferring terabytes of data between FOTOhub S3 and FOTOhub Compute instances is completely free and exceptionally fast, as traffic never leaves the datacenter.
Endpoint: https://s1.fotohub.app
Use Case: Storing thousands of generated images, maintaining a library of LoRA weights, or backing up processed datasets.

9.3 Bring Your Own Bucket (BYOB) Destinations

If your architecture relies heavily on external storage, you can stream outputs (rendered images, video edits, 3D files) directly from your compute instances into external buckets across AWS S3, Cloudflare R2, Google Cloud Storage, or Supabase. Check our Delivery to Your Bucket Guide for step-by-step connection tutorials.

10. Firecracker Sandboxes (Agent Compute)

For workloads requiring instantaneous startup and strict isolation—such as executing AI-generated Python code, evaluating agent steps, or running serverless functions—we provide highly optimized Firecracker microVMs.

10.1 Sandbox Architecture

Firecracker provides the strict security of hardware virtualization (KVM) combined with the speed and agility of container execution.

Cold Start Time: Sub-200 milliseconds.
Isolation: Full hardware-level kernel isolation per workload. No shared kernels.
Communication: Host-Guest communication is handled securely over vsock (Port 9999), bypassing standard networking stacks for improved latency and security.

10.2 API Endpoint & Sandbox Billing

You can execute arbitrary code inside a clean Python environment instantly:

http

POST https://apis.fotohub.app/sandbox/exec-python
Authorization: Bearer fh_live_YOUR_API_KEY

Sandbox Billing: Extremely cost-effective. Executions are billed strictly from your USD wallet at approximately $0.00008 per run, based precisely on the milliseconds of compute time used.

11. Common Use Cases and Reference Patterns

11.1 Large Language Model (LLM) Inference (vLLM / TGI)

Deploy models like Llama-3 (8B or 70B), Mistral, Qwen, or deepseek.

Hardware Profile: g5.xlarge (A10G) for 7B-14B models. g5.12xlarge (4x A10G) or g4dn.12xlarge (4x T4) for larger models or higher concurrency throughput.
Architectural Pattern:
1. Boot instance via API.
2. Attach a persistent EBS volume containing pre-downloaded weights mounted at /data/models.
3. Start the vLLM docker container exposing port 8000.
4. Use the Firewall API to open port 8000.
5. Map a custom domain via the Route53 DNS manager for easy access.

11.2 Image Generation (ComfyUI / Stable Diffusion / Flux)

Run automated render farms or host multi-user UI instances for designers.

Hardware Profile: g5.xlarge (A10G 24GB VRAM) is the gold standard for SDXL and Flux models requiring high VRAM. g4dn.xlarge (T4 16GB) is incredibly cost-effective for SD 1.5 pipelines.
Architectural Pattern:
1. Use a custom AMI pre-configured with CUDA drivers and ComfyUI.
2. Start instance on Spot pricing to maximize margins.
3. Queue API requests to the instance.
4. Stream generated images directly to FOTOhub S3 using BYOB Destinations.

11.3 Machine Learning Training & LoRA Fine-tuning

Hardware Profile: g5.2xlarge or g5.4xlarge.
Architectural Pattern:
1. Provision instance via API.
2. Attach high-IOPS io2 volume for rapid dataset streaming and checkpoint saving.
3. Execute training script.
4. Push final LoRA checkpoints to FOTOhub S3.
5. Instance Auto-Stops after max_runtime_hours to ensure you aren't billed if the script crashes.

11.4 CPU Batch Processing & Video Transcoding

Hardware Profile: c5.4xlarge (16 vCPU) or m5.4xlarge.
Architectural Pattern:
1. Spin up a cluster of 10-20 Spot instances via API.
2. Instances pull jobs from an SQS or RabbitMQ queue.
3. Process FFmpeg transcoding locally.
4. Upload finished assets to Cloudflare R2 via BYOB endpoints.
5. Terminate instance locally (sudo shutdown now) which signals the FOTOhub API to stop billing.

12. Frequently Asked Questions (FAQ)

1. Do you offer free credits for new users?

No. To maintain system stability, guarantee performance, and prevent widespread abuse (like crypto-mining botnets), FOTOhub Compute operates strictly on a prepaid USD system. You must add at least $0.50 from a valid payment method to begin provisioning.

2. Can I pay in EUR, PLN, or Crypto?

No. The FOTOhub wallet only accepts and denominates balances in United States Dollars (USD). We do not support local currencies like Polish Zloty (PLN) or Euros, nor do we accept cryptocurrency payments. All APIs and billing systems expect USD.

3. What happens if AWS terminates my Spot Instance?

Spot instances can be reclaimed by AWS if the market price spikes or capacity drops. If this occurs, FOTOhub will immediately emit a termination event via the SSE Event Bus. You will only be billed for the exact seconds the instance was running. We highly recommend designing your pipelines to be stateless and checkpointing models frequently to EBS or S3 to gracefully handle interruptions.

4. Can I SSH into a Firecracker Sandbox?

No. Sandboxes are entirely ephemeral and designed for single-execution API calls (/sandbox/exec-python). They are destroyed immediately after the code execution finishes. For interactive shell access, you must provision a standard EC2 instance like a T3 or G4dn.

5. How do I open port 80 or 443 for my web server?

By default, only port 22 is open to the internet. You must use the Security Groups API (POST /compute/v1/firewall/rules) to explicitly open port 80 (HTTP), port 443 (HTTPS), or any other custom port for your specific instance ID.

6. Are my EBS volumes deleted when I terminate an instance?

By default, root operating system volumes are deleted on instance termination, while secondary attached volumes (where you should store data) are retained. You can modify this behavior using the delete_on_termination flag during provisioning or volume attachment.

7. Can I reserve instances for long-term discounts?

Currently, FOTOhub Compute offers On-Demand and Spot pricing. Long-term reservations (1-year or 3-year commits) are handled on a case-by-case basis for Enterprise customers only. Please contact support if you are interested in enterprise commits.

8. What operating systems are supported?

We currently support highly optimized AMIs based on Ubuntu 22.04 LTS, Ubuntu 20.04 LTS, and Amazon Linux 2023. These AMIs come pre-installed with the necessary NVIDIA drivers and CUDA toolkits. Custom AMIs and Windows Server are not supported at this time.

9. Why can't I provision a `g5.48xlarge`?

Very large instance types have strict account-level limits to protect against accidental spending and capacity hoarding. If you require massive multi-GPU instances, please submit a quota increase request through the Console UI providing details of your workload.

10. How is network egress billed?

Data transferred out of FOTOhub to the public internet costs $0.09/GB. Data transferred internally to other FOTOhub instances or directly to FOTOhub S3 (s1.fotohub.app) within the Frankfurt region is entirely free ($0.00/GB).

11. Is there a cold start penalty for GPU instances?

Yes. Unlike Firecracker sandboxes which boot in milliseconds, standard EC2 GPU instances require time to allocate hardware, boot the OS, and initialize drivers. Expect a provisioning and boot time of roughly 60 to 120 seconds before the instance is reachable via SSH.

12. Can I change my instance type later?

No. Instance types are fixed at creation. If you need a more powerful instance, you should take a snapshot of your EBS volume, provision a new instance of the desired type, and attach the snapshot.

13. What's Next? Further Reading & API References

Ready to start building? Dive deeper into specific topics with our comprehensive sub-pages and API references:

GPU Rental & Instance Lifecycle: Detailed REST API references for starting, stopping, restarting, monitoring, and terminating EC2 instances programmatically.
Volumes & Persistent Storage: Learn how to create, attach, format, mount, and snapshot EBS gp3/io2 volumes to ensure data durability.
Networking, DNS & Security Groups: A deep dive into managing inbound/outbound rules, understanding VPC contexts, and securing your applications.
FOTOhub S3 Integration: Comprehensive guide on using the FOTOhub S3-compatible API for zero-egress data storage and retrieval.
BYOB Destinations: Step-by-step instructions for configuring external buckets (AWS, Cloudflare, GCS) for automatic data routing and output streaming.
Firecracker microVM Sandboxes: Full documentation on the sub-200ms Python execution environment, including limitations, package support, and payload formatting.
Wallet & Billing API: Learn how to programmatically check your USD balances, view transaction history, and handle low-balance webhook alerts in your application.

14. Troubleshooting & Best Practices

When building production-grade AI applications on FOTOhub Compute, you may encounter edge cases or operational challenges. This section provides detailed troubleshooting steps and architectural best practices.

14.1 Common Error Codes

402 Payment Required: Your USD wallet balance has fallen below the required $0.50 minimum to provision a new instance. Top up your wallet in the Console UI.
403 Forbidden: Your API key is invalid, revoked, or lacks the necessary permissions to execute this action. Check your API key prefix (ensure it starts with fh_live_).
404 Not Found: The requested instance ID or volume ID does not exist, or it has been permanently terminated and purged from the system.
429 Too Many Requests: You are hitting the API rate limit (default is 100 requests per minute). Implement exponential backoff in your client code.
503 Service Unavailable: The specific Availability Zone is experiencing a capacity constraint for the requested instance type (common with Spot GPUs during peak hours). Try provisioning in a different zone or using a different instance family.

14.2 Best Practices for Spot Instances

Spot instances offer incredible savings (up to 70%), but they come with the caveat of potential interruption.

Stateless Design: Architect your applications to be as stateless as possible. Store all critical data and model weights on attached EBS volumes, not the root volume.
Frequent Checkpointing: If you are running long training jobs, implement a checkpointing mechanism that saves progress to an EBS volume or FOTOhub S3 every few minutes.
Automated Retries: Wrap your provisioning logic in a retry block. If an instance is terminated by AWS, your system should automatically request a replacement.
Graceful Shutdown Handling: Listen to the SSE Event Bus for the spot_interruption_warning event. You typically have 2 minutes to save state and unmount drives before the instance is forcibly shut down.

14.3 Monitoring and Telemetry

While FOTOhub handles the underlying infrastructure, monitoring your application's health remains crucial.

CPU and RAM Utilization: Install htop or glances for real-time monitoring. For programmatic monitoring, install the Datadog or Prometheus agent.
GPU Utilization: Use nvidia-smi to monitor VRAM usage, GPU temperature, and power draw. For continuous monitoring, the dcgm-exporter is highly recommended for exporting metrics to Prometheus.
Log Management: Do not store massive log files on the root volume. Stream your logs directly to FOTOhub S3, CloudWatch, or a third-party service like Datadog to ensure they survive instance termination.

14.4 Performance Optimization Tips

Maximize Storage Throughput: If your workload is highly I/O bound (e.g., streaming massive datasets during training), always opt for io2 Provisioned IOPS volumes rather than gp3.
Pre-bake Custom AMIs: If you find yourself repeatedly installing the same dependencies (PyTorch, vLLM, ComfyUI nodes) on boot, consider creating a custom AMI snapshot. This will drastically reduce your instance boot time from several minutes to under 60 seconds.
Leverage FOTOhub S3: Avoid downloading large model weights from the public internet (HuggingFace, Civitai) on every boot. Download them once, store them in FOTOhub S3, and pull them across our ultra-fast internal network for $0.00 egress cost.

15. Compliance and Data Residency

FOTOhub Compute operates out of the AWS eu-central-1 (Frankfurt) region. This ensures strict compliance with European data protection regulations, including the GDPR.

By default, all data written to attached EBS volumes or FOTOhub S3 remains physically within the Frankfurt region. We do not replicate your data across regions without your explicit API request.

For enterprise customers requiring BAA (Business Associate Agreements) for HIPAA compliance, or specific SOC2 Type II reporting, please contact our enterprise sales team. Note that standard self-serve accounts do not include customized compliance reporting.

Cloud Compute & Sandboxes Architecture: The Complete Guide ​

1. Introduction: What is FOTOhub Compute and Who is it For? ​

1.1 The FOTOhub Compute Philosophy ​

1.2 Target Audience and Use Cases ​

AI/ML Engineers & Researchers ​

Autonomous Agent Developers ​

Data Engineers & Big Data Practitioners ​

Backend & Full-Stack Developers ​

1.3 Key Platform Capabilities ​

2. High-Level Architecture & Control Plane ​

2.1 The Compute Engine (EC2 Abstraction) ​

2.2 The Agent Compute Orchestrator (MicroVMs) ​

3. Comprehensive Hardware Specifications & 22-Instance Catalog ​

3.1 GPU Accelerated Instances (NVIDIA) ​

3.2 General Purpose CPU Instances (T3 & M5 Families) ​

3.3 Compute & Memory Optimized Instances (C5 & R5 Families) ​

4. Querying the Live Catalog API ​

Response Example ​

5. Infrastructure Comparison: FOTOhub vs Alternatives ​

5.1 The FOTOhub Advantage ​

6. Quick Start in 5 Minutes ​

6.1 Prerequisites ​

6.2 Step 1: Provision an Instance ​

6.3 Step 2: Retrieve Ephemeral SSH Key and Connect ​

7. Billing & USD Wallet Deep Dive ​

7.1 The Billing Lifecycle Flow ​

7.2 Key Billing Rules & Enforcements ​

8. Security Model Deep Dive ​

8.1 Multi-Layered Perimeter Defense ​

9. Integrated Storage & Data Pipelines ​

9.1 Elastic Block Store (EBS) Persistent Volumes ​

9.2 FOTOhub S3 Cloud Storage ​

9.3 Bring Your Own Bucket (BYOB) Destinations ​

10. Firecracker Sandboxes (Agent Compute) ​

10.1 Sandbox Architecture ​

10.2 API Endpoint & Sandbox Billing ​

11. Common Use Cases and Reference Patterns ​

11.1 Large Language Model (LLM) Inference (vLLM / TGI) ​

11.2 Image Generation (ComfyUI / Stable Diffusion / Flux) ​

11.3 Machine Learning Training & LoRA Fine-tuning ​

11.4 CPU Batch Processing & Video Transcoding ​

12. Frequently Asked Questions (FAQ) ​

1. Do you offer free credits for new users? ​

2. Can I pay in EUR, PLN, or Crypto? ​

3. What happens if AWS terminates my Spot Instance? ​

4. Can I SSH into a Firecracker Sandbox? ​

5. How do I open port 80 or 443 for my web server? ​

6. Are my EBS volumes deleted when I terminate an instance? ​

7. Can I reserve instances for long-term discounts? ​

8. What operating systems are supported? ​

9. Why can't I provision a g5.48xlarge? ​

10. How is network egress billed? ​

11. Is there a cold start penalty for GPU instances? ​

12. Can I change my instance type later? ​

13. What's Next? Further Reading & API References ​

14. Troubleshooting & Best Practices ​

14.1 Common Error Codes ​

14.2 Best Practices for Spot Instances ​

14.3 Monitoring and Telemetry ​

14.4 Performance Optimization Tips ​

15. Compliance and Data Residency ​

Cloud Compute & Sandboxes Architecture: The Complete Guide

1. Introduction: What is FOTOhub Compute and Who is it For?

1.1 The FOTOhub Compute Philosophy

1.2 Target Audience and Use Cases

AI/ML Engineers & Researchers

Autonomous Agent Developers

Data Engineers & Big Data Practitioners

Backend & Full-Stack Developers

1.3 Key Platform Capabilities

2. High-Level Architecture & Control Plane

2.1 The Compute Engine (EC2 Abstraction)

2.2 The Agent Compute Orchestrator (MicroVMs)

3. Comprehensive Hardware Specifications & 22-Instance Catalog

3.1 GPU Accelerated Instances (NVIDIA)

3.2 General Purpose CPU Instances (T3 & M5 Families)

3.3 Compute & Memory Optimized Instances (C5 & R5 Families)

4. Querying the Live Catalog API

Response Example

5. Infrastructure Comparison: FOTOhub vs Alternatives

5.1 The FOTOhub Advantage

6. Quick Start in 5 Minutes

6.1 Prerequisites

6.2 Step 1: Provision an Instance

6.3 Step 2: Retrieve Ephemeral SSH Key and Connect

7. Billing & USD Wallet Deep Dive

7.1 The Billing Lifecycle Flow

7.2 Key Billing Rules & Enforcements

8. Security Model Deep Dive

8.1 Multi-Layered Perimeter Defense

9. Integrated Storage & Data Pipelines

9.1 Elastic Block Store (EBS) Persistent Volumes

9.2 FOTOhub S3 Cloud Storage

9.3 Bring Your Own Bucket (BYOB) Destinations

10. Firecracker Sandboxes (Agent Compute)

10.1 Sandbox Architecture

10.2 API Endpoint & Sandbox Billing

11. Common Use Cases and Reference Patterns

11.1 Large Language Model (LLM) Inference (vLLM / TGI)

11.2 Image Generation (ComfyUI / Stable Diffusion / Flux)

11.3 Machine Learning Training & LoRA Fine-tuning

11.4 CPU Batch Processing & Video Transcoding

12. Frequently Asked Questions (FAQ)

1. Do you offer free credits for new users?

2. Can I pay in EUR, PLN, or Crypto?

3. What happens if AWS terminates my Spot Instance?

4. Can I SSH into a Firecracker Sandbox?

5. How do I open port 80 or 443 for my web server?

6. Are my EBS volumes deleted when I terminate an instance?

7. Can I reserve instances for long-term discounts?

8. What operating systems are supported?

9. Why can't I provision a `g5.48xlarge`?

10. How is network egress billed?

11. Is there a cold start penalty for GPU instances?

12. Can I change my instance type later?

13. What's Next? Further Reading & API References

14. Troubleshooting & Best Practices

14.1 Common Error Codes

14.2 Best Practices for Spot Instances

14.3 Monitoring and Telemetry

14.4 Performance Optimization Tips

15. Compliance and Data Residency