Dedicated server or VPS: which to choose for your business

This article is for a manager or administrator deciding where to host an accounting system, remote desktops, files or a website. After reading it you will know how a dedicated server differs from a VPS, what you are running on today and which type of server suits your workload.

What they are in plain terms#

A dedicated server is a separate physical computer in a data centre, rented in full by one company. Every processor core, all the memory and all the disks work only on your tasks, and the operating system is installed directly on the hardware.

A VPS (virtual private server; also called a VDS or a virtual machine) is a part of a physical server. That server runs a hypervisor — a program that divides the processor, memory and disks between the virtual machines of different customers. You have your own operating system with administrator rights, but the hardware underneath is shared.

A comparison on seven criteria#

What you need to make the choice#

• A list of the systems to be hosted: accounting system (BAS, M.E.Doc), remote desktops, files, website, mail.
• The number of people working at the same time, and the hours of heaviest load.
• The size of the databases and files now, and the growth you expect over a year.
• Administrator access to the current server — to measure rather than guess.

How to choose: five steps#

1. Find out what you are running on today. On Linux with systemd, run:

   systemd-detect-virt

   The answer none means a physical server; kvm, vmware, microsoft, xen and the like mean a virtual machine; lxc or openvz means a container. On Windows Server, open PowerShell and run:

   Get-CimInstance -ClassName Win32_ComputerSystem | Select-Object Manufacturer, Model, HypervisorPresent

   For a virtual machine the Model field holds a name such as Virtual Machine (Hyper-V), VMware Virtual Platform or VMware7,1, Standard PC (Q35 + ICH9, 2009) (QEMU/KVM), or a name set by the provider; for a physical server it holds the model of the board or of the server itself. The HypervisorPresent field is only a secondary sign: it can be True on a physical server too, where the Hyper-V role or virtualization-based security is enabled.

2. If it is a Linux virtual machine, check whether the neighbours get in the way. At the hours of heaviest load, run:

   vmstat 5 12

   The command prints a line every five seconds, twelve in total. The first line shows averages since the system booted — skip it. The st column shows the share of processor time the hypervisor took away from your machine, and wa shows time spent waiting for disk input/output. There is no single threshold, but if st stays at several per cent or more throughout working hours and people complain that everything is “slow”, the cause is most likely the shared hardware. Windows has no built-in counterpart — rely on the timing of a typical operation from the next step.

3. Write down how many resources you have and how many you use. On Linux:

   nproc
   free -h
   df -h

   The number of logical processors, the memory and the disk space in use are the lower bound for the new server. On Windows, Task Manager shows the processor and memory on the “Performance” tab, and the PowerShell command Get-PSDrive -PSProvider FileSystem shows used and free disk space. Also time one heavy typical operation — posting a month of documents, building a large report, making a database backup — in the morning and at peak hours. You will need these numbers for comparison after the move.

4. Answer five questions.

   • Is the load constant — do people work with the server all day, every working day?
   • Is there an accounting database or another database where every second of waiting is noticed?
   • Do more than a few people work on the server over RDP at the same time?
   • Do you need several separate virtual machines for different roles?
   • Does the security policy of your company or of your customer require that no one else’s systems share the hardware?

   Two or more “yes” answers are a strong argument for a dedicated server. All “no” — a VPS will most likely be enough.

5. Choose the line and the configuration. How to do that is described in “How to choose a server line and configuration”; for accounting databases there is a separate calculation — “A server for BAS for 5, 10, 20 and 50 users”.

When you need a dedicated server#

• Constant load. Shared hardware pays off when resources sit idle most of the time. If the server is busy all day, every working day, there is no point in sharing it.
• Accounting databases and other databases. Accounting systems such as BAS depend first of all on the speed of a single core, not on the number of cores; a database needs a fast disk with predictable latency. On a dedicated server you know the processor model and the disk type, and you share them with nobody. More in “A server for PostgreSQL and MS SQL Server”.
• Many RDP users. Every remote desktop session consumes memory and processor time. With dozens of sessions, a one-second delay is felt by everyone at once.
• Your own hypervisor. You can divide a dedicated server into virtual machines yourself: one for the database, one for the terminal server, one for a test environment. That is the flexibility of a VPS without outside neighbours — see “Proxmox VE on a dedicated server”.
• Isolation requirements. If your internal security policy or an agreement with a customer requires that data not sit on the same hardware as other parties’ systems, a physical server meets that requirement.

When a VPS is enough#

• A small website or landing page, an auxiliary service with a light load.
• Tests. Check an update, set up a test environment, show it to a contractor — and switch it off.
• Short projects lasting a few weeks, when it matters to get a server in minutes and to give it up just as quickly.

The common feature: the load is light or uneven, and a brief slowdown harms nobody.

What United Cloud offers#

To be frank: at the moment we offer dedicated servers only. Virtual servers are coming soon. If a VPS really is enough for your task, you do not have to take a dedicated server for it; for tests and non-critical tasks we have the budget Lite line.

Every dedicated server includes:

• network bandwidth and unmetered traffic;
• network-level DDoS protection (L3/L4): it switches on automatically when an attack is detected, with no extra charge for the volume or duration of the attack. It does not filter application-level (L7) attacks — see “DDoS protection: what it filters and what it does not”;
• the whole server at your disposal.

On the main lines you also get a private network between servers, including between countries, and backup space on separate storage — 500 GB, with the exact amount shown on the server card. The Lite line has no private network, backup space or remote console, so for production systems we recommend Standard or Business.

What remains your job#

• Administering the operating system: updates, users, the firewall, service settings. Where to begin is covered in the first-hour articles for Linux and Windows Server.
• Backups. On the main lines the space is there, but copying is not automatic: you set up the schedule and the restore test yourself. The scheme is in “The 3-2-1 rule”.
• Watching the disks and RAID. Faulty hardware is replaced in the data centre on a request made through support, but noticing that a disk has dropped out of the array is up to you — see “How to check software RAID and disk health”.

We do not have a customer account or a control panel yet. To power-cycle the server, reinstall the operating system or get the remote console, contact support: phone +38 044 206 08 08, email info@united.net.ua. Technical support works around the clock. Software licences are not covered in this article.

How to check the result#

1. systemd-detect-virt answers none, and on Windows the Model field shows real hardware. If you have installed your own hypervisor, check on the hypervisor itself, not inside a virtual machine.

2. The processor and disks match the configuration from the catalogue:

   lscpu
   lsblk -d -o NAME,SIZE,ROTA,MODEL

   In the ROTA column, 0 means an SSD or NVMe drive and 1 means a hard disk. Behind a hardware RAID controller the command shows the logical volume, not the disks themselves.

3. In the output of vmstat 5 12 the st column is zero under any load.
4. The heavy operation you timed in step three runs no slower than before, and takes the same time in the morning and at peak hours. Check the response time from the office with “How to measure response time and the route to a server”.

Common mistakes#

• Comparing only the number of cores and gigabytes. A virtual core is usually a thread of a shared processor, not a physical core: 8 vCPU on a VPS and 8 cores of a dedicated server are different things. Look at the processor model and the disk type.
• Assuming a dedicated server is fault-tolerant by itself. One server is one point of failure. Critical systems need backups and a second server in another country: “A standby site: three schemes and what to check”.
• Confusing RAID with a backup. A mirror saves you from a failed disk, but not from a database deleted by mistake or from ransomware.
• Choosing by the name instead of by measurements. If st on your current VPS is zero and the system is slow anyway, the cause may be the database settings or a shortage of memory — moving to a dedicated server will not remove it by itself.

What next#

• Dedicated server catalogue — current availability in each country and the delivery time.
• Servers for business — typical tasks and the lines we recommend for them.
• How to choose a country for the server and the “Locations” page.
• From request to access: availability and delivery times.
• A checklist for moving from a Ukrainian data centre to Europe.