Work Server

The Atto network operates without transaction fees. Instead, to prevent spam and prioritize network resources, it requires a small amount of Proof of Work (PoW) to be computed for each transaction before it's submitted. The Work Server is a dedicated service designed to perform these PoW computations.

Clients and applications wishing to send transactions on the Atto network will typically delegate the PoW calculation to a Work Server.

Why is a Work Server Required?

Fee-less Transactions: Atto's design replaces monetary fees with computational work. This PoW needs to be generated for every transaction.
Resource Intensive: While the PoW difficulty is generally low, it can still be resource-intensive, especially for devices with limited processing power or when submitting many transactions.
Centralized Computation (Optional): Deploying a dedicated Work Server (or multiple) allows various applications and services to offload this computational task, streamlining their own operations.

Performance Considerations

The time taken to compute the PoW can vary significantly based on the available hardware:

CPU: Standard CPUs can compute the PoW, but it might take a noticeable amount of time, potentially several minutes for a single transaction depending on the CPU's performance. This might be acceptable for development or low-throughput scenarios.
GPU (Recommended for Live Environments): For production or high-throughput environments, using GPUs is highly recommended. GPUs can perform the PoW calculations much faster.
- For instance, an nvidia-l4 GPU can typically calculate the required PoW in approximately 1 second.
- Using a GPU significantly improves the user experience and the efficiency of submitting transactions.

Community Support: If running and maintaining a dedicated GPU-accelerated Work Server is a barrier for your project, please reach out to the Atto team. We are open to discussing options for sharing access to our Work Server resources to facilitate adoption and development on the Atto network.

Docker Images Available

The Work Server is available in two main image variants, typically hosted on ghcr.io/attocash/work-server:

ghcr.io/attocash/work-server:cpu: Optimized for CPU-based PoW calculation.
ghcr.io/attocash/work-server:cuda: Optimized for GPU-based PoW calculation using NVIDIA CUDA. This image requires a compatible NVIDIA GPU and drivers on the host machine/node. *ghcr.io/attocash/work-server:rocm: Optimized for GPU-based PoW calculation using AMD ROCM. This image requires a compatible AMD GPU and drivers on the host machine/node.

Quick Reference

Ports

Port	Purpose	Exposure
8080	REST API	Cluster‑internal only.
8081	Liveness `/health` & metrics `/prometheus`	Cluster‑internal only.

Kubernetes Examples

Below are examples of how to deploy a Work Server on Kubernetes.

CPU-based

This manifest deploys a Work Server using the CPU image. It's suitable for development, testing, or low-traffic scenarios.

View Details

apiVersion: apps/v1
kind: Deployment
metadata:
  name: work-server-cpu
  namespace: work # Example namespace
  labels:
    app: work-server-cpu
spec:
  replicas: 1
  selector:
    matchLabels:
      app: work-server-cpu
  template:
    metadata:
      labels:
        app: work-server-cpu
    spec:
      containers:
        - name: work-server
          image: ghcr.io/attocash/work-server:cpu # CPU image
          imagePullPolicy: Always
          ports:
            - name: http
              containerPort: 8080
            - name: management
              containerPort: 8081
          resources:
            requests:
              cpu: "2" # Adjust based on expected load
              memory: "2Gi"
            limits:
              cpu: "2" # Adjust based on expected load
              memory: "2Gi"
              ephemeral-storage: "1Gi"
          startupProbe:
            httpGet:
              path: /health
              port: health # Refers to named port 'health' (8081)
            failureThreshold: 180
            periodSeconds: 1
          livenessProbe:
            httpGet:
              path: /health
              port: health # Refers to named port 'health' (8081)
            failureThreshold: 5
            periodSeconds: 60

GPU-accelerated

View Details

This manifest deploys a Work Server using the CUDA image and targets a node with an NVIDIA L4 GPU. This is recommended for live/production environments.

Prerequisites:

Your Kubernetes nodes must have NVIDIA GPUs installed.
NVIDIA device drivers and the Kubernetes device plugin for NVIDIA GPUs must be configured on your cluster.
The example uses node selectors common in GKE for GPU resources. Adapt these to your cluster's specific configuration (e.g., for other cloud providers or on-premise setups).

apiVersion: apps/v1
kind: Deployment
metadata:
  name: work-server-gpu
  namespace: work # Example namespace
  labels:
    app: work-server-gpu
spec:
  replicas: 1
  selector:
    matchLabels:
      app: work-server-gpu
  template:
    metadata:
      labels:
        app: work-server-gpu
    spec:
      # Node selectors are crucial for scheduling pods on GPU-enabled nodes.
      # The specific selectors depend on your Kubernetes distribution and GPU setup.
      nodeSelector:
        cloud.google.com/gke-accelerator: nvidia-l4 # Example for GKE L4 GPU
        # For other environments, or more specific scheduling, adjust accordingly.
        # e.g., cloud.google.com/compute-class: "Accelerator"
        # If using reservations (as in the Terraform example):
        # cloud.google.com/reservation-name: "work-server-l4" # Replace with your reservation name
        # cloud.google.com/reservation-affinity: "specific"
      tolerations: # Often needed if GPU nodes have taints
        - key: "nvidia.com/gpu"
          operator: "Exists"
          effect: "NoSchedule" # Or "PreferNoSchedule"
      containers:
        - name: work-server
          image: ghcr.io/attocash/work-server:cuda # CUDA image for GPU
          imagePullPolicy: Always
          ports:
            - name: http
              containerPort: 8080
            - name: health
              containerPort: 8081
          resources:
            limits:
              nvidia.com/gpu: 1 # Request 1 NVIDIA GPU
              # memory: "4Gi" # Adjust memory as needed for GPU workloads
              # cpu: "1"      # Adjust CPU as needed
            requests:
              nvidia.com/gpu: 1 # Request 1 NVIDIA GPU
              # memory: "4Gi"
              # cpu: "1"
          startupProbe:
            httpGet:
              path: /health
              port: health # Refers to named port 'health' (8081)
            failureThreshold: 180
            periodSeconds: 1
          livenessProbe:
            httpGet:
              path: /health
              port: health # Refers to named port 'health' (8081)
            failureThreshold: 5
            periodSeconds: 60

Note on GPU Deployment:

The nodeSelector and tolerations are critical for ensuring the pod is scheduled on a node with the required GPU resources. The exact labels and taints depend on how your Kubernetes cluster's GPU support is configured (e.g., nvidia.com/gpu is common).
The resources.limits and resources.requests for nvidia.com/gpu: 1 ensure the pod is allocated one GPU. Adjust CPU and memory resources as needed.
Ensure the ghcr.io/attocash/work-server:cuda image is compatible with the CUDA drivers installed on your GPU nodes.