Server rental store

AI in Tyne and Wear

AI in Tyne and Wear: Server Configuration for Regional Deployment

This article details the server configuration recommended for deploying and supporting Artificial Intelligence (AI) applications within the Tyne and Wear region. This guide is intended for newcomers to our MediaWiki site and focuses on practical, scalable solutions. We aim to provide a robust and efficient infrastructure capable of handling the demands of various AI workloads, from machine learning model training to real-time inference. Consideration is given to cost-effectiveness, maintainability, and future scalability.

Overview

The deployment strategy centres around a hybrid approach, leveraging both on-premise infrastructure for sensitive data and cloud-based resources for burstable workloads. This allows for flexibility and control while minimizing capital expenditure. The core on-premise system will reside within a secure data centre in Newcastle upon Tyne, with connectivity to major cloud providers (AWS, Azure, and Google Cloud Platform) via dedicated high-bandwidth links. This article will focus on the on-premise requirements, with notes on cloud integration. We will cover server specifications, networking, storage, and software stacks. See also: Data Centre Security Protocol and Network Topology Overview.

Hardware Specifications

The following table outlines the recommended hardware specifications for the core on-premise AI server cluster. The cluster will be divided into three tiers: Data Ingestion, Model Training, and Inference.

Tier Server Role CPU RAM GPU Storage Network Interface
Data Ingestion Data Collection & Preprocessing 2 x Intel Xeon Gold 6338 256GB DDR4 ECC None 4 x 4TB NVMe SSD (RAID 10) 10GbE
Model Training Machine Learning Model Training 2 x AMD EPYC 7763 512GB DDR4 ECC 4 x NVIDIA A100 80GB 8 x 8TB SAS HDD (RAID 6) + 2 x 1TB NVMe SSD (OS) 100GbE
Inference Real-Time Prediction and Analysis 2 x Intel Xeon Silver 4310 128GB DDR4 ECC 2 x NVIDIA T4 16GB 2 x 2TB NVMe SSD (RAID 1) 25GbE

These specifications are a starting point and can be adjusted based on the specific AI models and datasets being utilized. Refer to Hardware Procurement Guidelines for approved vendor lists. Regular monitoring of resource utilization is essential; see Server Monitoring Dashboard.

Networking Infrastructure

A robust network is critical for AI workloads, particularly for data transfer and distributed training. The following table details the networking requirements.

Component Specification Purpose
Core Switches Arista 7050X Series (x2) High-speed switching and routing
Server Network Adapters As specified in Hardware Specifications Connectivity to the network
Inter-Server Links 100GbE QSFP28 High-bandwidth communication between training servers
Internet Connectivity 10Gbps Dedicated Link Access to external resources and cloud providers
Firewall Palo Alto Networks PA-820 Network security and access control

Network segmentation is crucial for security. Separate VLANs should be configured for each tier (Data Ingestion, Training, Inference) and for management traffic. Details can be found in the Network Security Policy. Latency between servers should be minimized, ideally under 1ms.

Storage Configuration

Data storage needs vary greatly depending on the application. The on-premise storage solution is designed to handle both structured and unstructured data.

Storage Type Capacity Technology Purpose
Raw Data Storage 200TB SAS HDD (RAID 6) Long-term storage of raw datasets
Model Storage 50TB NVMe SSD Fast access to trained models for inference
Temporary Storage 20TB per server NVMe SSD Temporary files during data processing and training
Backup Storage 500TB Tape Library (LTO-9) Offsite data backup and disaster recovery

Data lifecycle management policies should be implemented to ensure efficient storage utilization. See Data Backup and Recovery Procedures for detailed instructions. The storage system utilizes a distributed file system (Ceph) for scalability and redundancy.

Software Stack

The software stack will be based on Ubuntu Server 22.04 LTS. The key components are:

⚠️ *Note: All benchmark scores are approximate and may vary based on configuration. Server availability subject to stock.* ⚠️

View full article | Back to Server rental store