Skip to main content
Legia

IT Infrastructure & Cloud

Sizing a Server for Internal Applications: CPU, RAM, and Storage

The server specification for an internal application should be calculated from vendor requirements and real usage data, not guessed from headcount. This guide explains how, and when a VPS, a dedicated server, or a cloud instance fits better.

SEO and Content Specialist

8 min read

A dark data center corridor with glass walls

Short answer: the server specification for an internal application cannot be guessed from headcount. Use the minimum requirements of the operating system, database, and application vendors as the floor, then measure the real load: users active at the same time during peak hours, database size and growth, and CPU, RAM, and disk usage over at least one full business cycle. Choose a VPS or cloud instance if the size still needs adjusting, and a dedicated server if you need consistent performance or licenses tied to physical hardware.

The question of server specifications for internal applications usually comes up when a company installs a new system (ERP, HRIS, attendance app, warehouse system) or when an old application starts slowing down at month end. Application vendors often just write “minimum 16 GB RAM” with no explanation, and the procurement team needs a number it can justify. This article covers how to calculate it. Where to place the server (in the office, in the cloud, or on shared hosting) is covered in the article Office Server, Cloud, or Shared Hosting.

Why there is no “this many users, this many GB of RAM” formula

Server requirements are determined by how the application is used, not by the number of accounts. A hundred users who occasionally open a form are far lighter than ten users pulling a full-year report at the same time. That is why Microsoft, in its Windows Server hardware requirements documentation, states that it is unrealistic to give generally applicable recommended hardware requirements because deployments vary so widely. Microsoft instead recommends test deployments to determine the appropriate requirements.

The “users versus RAM” tables circulating online do not know which application you use, how large your database is, or when your peak hours are. The safer approach is the following four steps.

Illustration: server

Step 1: start from the vendors’ minimum requirements

Minimum requirements are the floor for the software to install and run, not a sizing for your workload. If the operating system, database, and application sit on one server, their requirements add up. Examples from official vendor documentation:

ComponentVendor minimumVendor note
Windows Server 20251.4 GHz 64-bit processor, 2 GB RAM, 32 GB disk space4 GB RAM recommended for Server with Desktop Experience; 32 GB is called an absolute minimum
SQL Server 20221 GB RAM (editions other than Express), 1.4 GHz x64 processor, 6 GB disk spaceRecommended RAM is at least 4 GB and should be increased as the database grows
PostgreSQL 18No minimum table, but gives a starting point for memory settingsFor a dedicated database server with 1 GB of RAM or more, shared_buffers at 25% of RAM is a reasonable starting value
Your business applicationAsk its vendor for the system requirements documentAlso ask what the numbers are based on: how many active users and how much data

One thing to remember from that table: Microsoft writes that actual requirements depend on the configuration, applications, and features installed. Minimum figures are only a starting point.

Step 2: calculate the load from usage patterns

Data to collect before deciding the server specification for an internal application:

  1. Users active at the same time during peak hours, not the number of accounts. Identify the peak moments: morning attendance, month-end closing, or payroll.
  2. Heavy jobs: large reports, bulk exports and imports, synchronization with other systems, and scheduled tasks at night. Jobs like these usually consume the most CPU and disk.
  3. Current database size and its monthly growth, taken from history, not estimates.
  4. Plans for the next few years: new branches, new modules, or integrations to be added.
  5. Other components on the same server: antivirus, monitoring agents, and backup software also use resources.

If the application already runs on an old server, this data can be read directly from that server. If the application is new, ask the vendor to run a load test in a test environment using your peak-hour scenario.

Step 3: measure first, then upsize

Raising specifications without monitoring data often just moves the problem. Three things to watch when reading usage figures:

  • Look at peaks, not averages. The AWS Well-Architected guidance recommends using the maximum or the 99th percentile instead of the average, so the figures reflect the user experience.
  • Observe over a full cycle. The same guidance gives the example that a two-week analysis can miss a monthly spike, leaving the server under-provisioned. For accounting or payroll applications, the cycle is at least one month.
  • RAM that looks full is not necessarily short. On Linux, the available column of the free command already accounts for the page cache, so read that column, not free. Databases also use memory deliberately: SQL Server will claim memory up to the max server memory limit that is set. A more reliable sign of insufficient RAM is persistent swapping or paging.

For CPU, check whether the process queue gets long during peak hours. For disk, look not only at the remaining space but also at read and write wait times, because a slow database is often caused by the disk, not the CPU.

Step 4: calculate storage needs

Storage is calculated from several parts, not just today’s database size:

  1. The operating system and applications, per the vendors’ requirements.
  2. The current database plus monthly growth multiplied by the number of months planned.
  3. Transaction logs and application logs, which can grow quickly if they are not rotated.
  4. User-uploaded files, such as attachments and scanned documents.
  5. Local backup working space, if backup copies are created on the same server before being sent elsewhere.

Leave free space agreed by the team so that warnings appear well before the disk fills up. Backups should not be kept only on the same server. The storage pattern is covered in the article The 3-2-1 Backup Strategy for Company Servers.

VPS, dedicated server, or cloud instance

Once the numbers are in, choosing the type of server becomes easier. A VPS is a virtual machine with a fixed plan size that shares physical hardware with other tenants. A dedicated server is a single physical machine used by you alone. A cloud instance is a virtual machine at a cloud provider whose type can be changed and which is generally billed by usage.

AspectVPSDedicated serverCloud instance
Physical hardwareShared with other tenantsEntirely yoursGenerally shared, with dedicated hardware options
Changing sizeMove to another plan at the same providerAdd or replace physical componentsChange the instance type
PerformanceCan be affected by other tenants’ load, depending on the providerMost consistentDepends on the instance type chosen
Per-core or per-socket licensingCheck with the licensorEasiest to countDedicated host options exist, for example AWS Dedicated Hosts, which support per-socket and per-core licenses
SuitsSmall to mid-sized internal applicationsHeavy, stable loads, or dedicated hardware requirementsLoads that fluctuate or are still being measured

For a new application whose load is not yet known, starting with a VPS or cloud instance is safer: the size can be raised once monitoring data has been collected. If months of monitoring show a heavy, stable load, a dedicated server or an office server is worth re-costing. AWS’s own guidance says right-sizing can be an iterative process triggered by changes in usage patterns.

Common mistakes

  • Buying specifications from a vendor brochure without asking the assumptions behind the figures.
  • Adding RAM because the memory graph looks full, when what is in use is cache.
  • Putting the application, database, and backups on one disk, so a single failure takes out all three.
  • Not installing monitoring, so the server is only found to be full when the application stops.
  • Forgetting to account for the growth of logs and uploaded files.

How Legia does it

In the Server Installation & Management service, Legia’s work starts with a 3-7 working day infrastructure audit to check the server’s condition, security configuration, and the backups in place. Legia handles physical servers in the office, virtual servers, and cloud, on both Linux and Windows Server, then sets up automated monitoring with early alerts. Under monthly management, the monthly report covers server availability, incidents, security updates installed, and resource usage, so decisions to raise or lower specifications are based on data. Server or cloud bills are still paid directly in your company’s name.

Frequently asked questions

How much RAM do 50 users need?

The number of users alone is not enough to answer that. Ask the application vendor for the requirement per active user and the assumed data size, then confirm with a load test or monitoring data from a server that is already running.

Is it better to buy a large specification from the start?

For a physical server, buying with room to grow for a few years makes sense, because adding components means buying and installing more hardware. For a VPS or cloud, starting from a measured size and raising it based on data is usually more economical.

Should the application and database be split onto separate servers?

It is worth considering if the two compete for memory or CPU during peak hours. If they stay on one server, limit the database’s memory. Microsoft, for example, recommends setting max server memory in SQL Server so the operating system and other applications do not run short of memory.

When should we move from a VPS to a dedicated server?

When monitoring shows inconsistent performance even after the size has been raised, when software licenses are counted per physical core, or when a requirement mandates dedicated hardware.

Sources

Other infrastructure services, including cloud migration and office network installation, are on the IT Infrastructure & Cloud page.

Category IT Infrastructure & Cloud

Keep reading

Related articles

All Articles