New Server Best Practices

Planning and Deploying Servers at CSAIL

This page helps you determine whether your research group needs a dedicated server and guides you through the deployment process.

Do You Actually Need a Dedicated Server?

Before buying or deploying hardware, consider CSAIL-wide resources that may meet your needs without the management overhead:

For web applications, databases, and general services:

• CSAIL shared web servers - Handle wikis, registration forms, content management, and custom domains with automatic HTTPS
• Shared database servers - MySQL, PostgreSQL available on-demand

For compute-intensive research:

• SLURM compute cluster - GPU and CPU resource allocation with job scheduling. Ideal for batch processing, machine learning training, and parallel workloads
• OpenStack private cloud - Self-service VMs for flexible, on-demand computing

For large datasets:

• CSAIL AFS storage - Shared, backed-up home directories and group storage
• NFS datasets - For public or semi-public data distribution

Consider a dedicated server only if:

• You need extreme performance isolation or custom hardware (specialized GPUs, specific CPU configurations)
• Your workload requires real-time, low-latency responsiveness
• You need complete OS-level control (custom kernels, specific driver versions, unusual configurations)
• You have heavy, sustained compute needs that exceed shared cluster availability

Server Purchasing and Deployment

Contact TIG before making any hardware purchases. We will help you determine the right approach and facility location based on your requirements, timeline, and current capacity.

TIG actively manages server procurement to give you access to competitive pricing and enterprise support:

• Shop your requirements across Tier-1 vendors
• Leverage CSAIL’s collective purchasing power for better rates
• Coordinate quarterly and bi-annual lab-wide purchases
• Handle procurement, deployment, racking, and warranty management
• Integrate new systems into the SLURM cluster immediately

Best Practices for Deployed Servers

If you’ve decided you need a dedicated server, follow these practices:

Hostnames

Choose clear, pattern-matchable hostnames that indicate the machine’s purpose and owning group:

Good: xyz-gpu01, xyz-gpu02 (for GPU cluster in XYZ group)
Avoid: mercury, venus, pluto (group affiliation unclear, hard to pattern-match)

Benefits:

• Makes configuration management easier for TIG
• Helps TIG identify the responsible contact years later when you’ve left CSAIL
• The initial registrant in WebDNS is recorded; ideally this is the primary user or PI

Email Configuration

Arrange for root email to reach someone in your group with management responsibility for the machine.

On CSAIL Linux:

Edit or create /etc/facter/facts.d/local.txt and add:

root_alias=yourname@csail.mit.edu

Changes take effect within 30 minutes.

On legacy CSAIL Debian:

Edit the root: line in /etc/aliases. For example:

root: tig-server-root@csail.mit.edu, yourname@csail.mit.edu

Remote Management

For servers requiring remote access, configure IPMI (Intelligent Platform Management Interface) for out-of-band console and reboot capability. TIG will set this up during deployment.

Security and Responsibility

Next Steps

1. Determine if you need dedicated hardware - Review options above
2. Contact TIG early - Email help@csail.mit.edu with your requirements, timeline, and any constraints
3. Work with TIG on procurement - If purchasing, TIG will handle vendor coordination and pricing
4. TIG coordinates deployment - We determine facility location and handle initial setup
5. You maintain the system - Patching, monitoring, and day-to-day management

See Also

• Server Purchasing Support and Strategy
• CSAIL Linux
• SLURM Compute Cluster
• OpenStack
• Data Security
• WebDNS

Questions? Email help@csail.mit.edu