What is a VPS and how does it work?

What is a VPS?

A VPS (Virtual Private Server) means splitting one physical server, via virtualization, into several independently running «virtual servers». Each VPS has its own OS, CPU/RAM/disk resources and full root access, as if it were a separate physical machine.

The idea started in 1972 at IBM: the VM/370 system could «share» one limited (and very expensive) mainframe between multiple users. 50 years later the concept has barely changed, but speed, isolation quality, and hardware capabilities reached a cosmic level.

Today VPS works like this: a KVM hypervisor runs on a powerful physical host, partitions resources and gives each customer their virtual machine. Every VPS has its own BIOS, vCPU, RAM, disk and network card; the user feels they got a real physical computer.

How does a VPS work?

KVM is part of the Linux kernel, merged into the Linux mainline in 2007. Today Google Cloud, AWS Nitro (Amazon’s custom version), DigitalOcean, Linode, Hetzner and nearly every major hosting provider run on KVM. We do too.

Technical view: Linux + KVM runs on the physical server. KVM turns CPU cores into vCPUs, splits RAM into blocks, the disk into LVM volumes, and the network into virtual bridges. Each VPS appears as an independent computer with its own BIOS, vCPU, RAM and disk. Inside you can install Ubuntu, Debian, AlmaLinux, Rocky Linux, Windows Server, or one of 400+ ready templates.

Isolation: how VPS, shared and containers differ

Isolation is the core question of the virtualization industry. Three different approaches exist, each with a different security level.

VPS (hardware virtualization, KVM). Isolation is enforced by the CPU itself through Intel VT-x / AMD-V. Each VPS boots its own kernel; one VPS cannot reach another’s memory because the chip blocks it physically. You upgrade the Linux kernel, install custom kernel modules, tune sysctls as you wish, none of it touches your neighbours.

Shared hosting + CloudLinux + KernelCare. To be honest: on modern shared hosting (like ours), CloudLinux LVE separates customers via Linux cgroups, and for everyday workloads it is strong enough. Changing the kernel, tweaking sysctls, loading kernel modules is then our responsibility, not yours. There is a subtle nuance: we use KernelCare (a CloudLinux Group product), and on every critical Linux kernel CVE we apply the patch without reboot, in real time. So our shared servers are always up to date and there is no planned kernel downtime. The difference remains: on a VPS you own the kernel, on shared we do, but always patched.

Container virtualization (Docker, LXC, OpenVZ). Here isolation is at the OS level: one Linux kernel is shared across all containers, each container running in a cgroup. Cheaper, less resource overhead. But the security record raises eyebrows. Notable CVEs:

• CVE-2019-5736 (runc/Docker): container escape and overwrite of the host runc binary. Critical.
• CVE-2022-0492 (cgroups v1): root escape, full container-to-host breakout.
• CVE-2024-21626 (runc): again, host file-write capability.
• Every few months a new «container escape» CVE appears.

There’s another nuance with containers: the host’s root user sees and can read all container files directly. With KVM that is impossible: VPS files are stored as a RAW disk image, opening and reading them requires a dedicated operation, and it is tracked in monitoring.

Containers are a great tool for packaging your own applications (Docker is the world standard for that). But for isolation between customers, hardware virtualization (KVM) is safer.

Epro.io note. We don’t sell «cheap container-based VPS». All our VPS and VDS plans use full hardware virtualization (KVM). That is a deliberate choice: real isolation between customers is only possible at the hardware level.

VPS types: KVM, VMware, Proxmox and others

Three main players compete in the industry: KVM, VMware, Hyper-V. Smaller but interesting options sit next to them: Xen, OpenVZ, LXC.

KVM (Kernel-based Virtual Machine) is open-source virtualization built directly into the Linux kernel. Free, flexible, universal: any OS can run on it. Google, AWS Nitro, Linode, DigitalOcean and Epro.io all chose KVM. It is the industry standard today.

VMware ESXi is a commercial enterprise platform. Very powerful, very expensive. In 2024 Broadcom bought VMware for $69 billion and raised license prices by 300%+. That became the biggest story in the virtualization world: thousands of companies started leaving the old platform.

Proxmox VE became a star after the Broadcom event. It is an open-source platform built on KVM, with a GUI. Developed in Austria since 2008. In 2024–2025 tens of thousands of enterprise customers migrated from VMware to Proxmox: comparable functionality, and the price is zero.

Epro.io note — VMware migration. Proxmox 8.0+ supports direct virtual machine import from VMware vCenter. We recently migrated several VMware environments into our infrastructure to KVM with minimal downtime. If you’re planning to leave VMware, trust a professional team and relax, we know the path.

Hyper-V is Microsoft’s virtualization solution on Windows Server. Still seen in Windows-oriented infrastructures, but gradually replaced by Linux + KVM.

OpenVZ and LXC are container-based technologies (mentioned above). Cheaper but with weaker isolation.

Key VPS advantages

• Full root access. Install any software, panel, or configuration you want.
• Guaranteed resources. CPU and RAM do not depend on neighbours.
• Hardware-level isolation. Separate environment, separate IP, separate OS.
• Scalable. When workload grows, you scale resources; no migration required.
• Enterprise hardware. RAID10 Enterprise NVMe, ECC RAM, redundant network.

Practical outcome: even if a neighbour site is hammered, yours stays snappy. You install Docker, Node.js, PostgreSQL, Redis, MongoDB. When workload grows, CPU/RAM go up within an hour. 99.9% SLA and local support are on your side.

VPS resources, transparent: CPU, RAM, IOPS, Bandwidth

A professional VPS provider should openly show how much resource you get. That is the industry transparency signal.

vCPU (virtual core). On a regular VPS, 1 vCPU is a slice of one physical hyper-thread, base clock ~2.5–3.5 GHz. On the VDS line, a vCPU is a dedicated physical core (Dedicated Core, no overcommit, guaranteed performance).

RAM. A non-shared, ECC-protected RAM block. We do not use memory ballooning or overcommit: a 4 GB plan means a real 4 GB. When RAM fills up, performance does not collapse.

NVMe SSD. Enterprise-class NVMe in a RAID10 configuration. Orders of magnitude faster than ordinary SATA SSD, with DataCenter-grade endurance.

IOPS (input/output operations per second). How many operations per second the disk can perform. Our Enterprise NVMe drives in RAID10 deliver very high IOPS — suitable for heavy databases and e-commerce. That is many times higher than what you see on shared hosting.

Bandwidth. Network speed depends on the plan: 100 Mbit/s on the VPS line, 250 Mbit/s to 1 Gbit/s on the upper VDS plans. Higher port speed is available on request for any plan. Traffic is also per-plan (1 TB to 30 TB); we don’t write «unlimited» because showing concrete numbers is more honest.

IPv4 / IPv6. Every plan includes one IPv4 and IPv6. Extra IPv4 on request.

CPU steal time. CPU time «stolen» by the hypervisor (a consequence of overcommit). Our monitoring tracks steal time; on VDS it is 0 because of the Dedicated CPU.

Plan migration: moving to a stronger plan

As your project grows, resources scale up. In our infrastructure, plan migration is technically simple:

• Upgrade to a stronger VPS. Within the same line (VPS-XS → VPS-S → VPS-M → ...) a quick power-cycle is enough: a few seconds of downtime.
• VPS to VDS (or back). Same flow: power-cycle + new resource profile, a few seconds of downtime.
• Disk grow. Done online, downtime is virtually zero.
• Disk shrink. Trickier: needs file system shrink + partition resize + image change. Possible, but we schedule a planned maintenance window.
• Move to a dedicated server. Migration to a new server, planned, with the support team.

Epro.io note — Private VLAN. If you need a private network between virtual machines (for example, an isolated high-speed link between a web server and a database), we offer Private VLAN: 1 Gbit/s = 10 AZN/month, 10 Gbit/s = 20 AZN/month. Important detail: this VLAN can connect not only VPSs but also VDSs and physical (dedicated) servers. So a hybrid infrastructure (e.g., application VMs + a dedicated database or 1C server) works in the same private network. Easy to order, infrastructure is ready.

When do you need a VPS?

When a site outgrows shared hosting, or you need stable performance, your own stack (Docker, Node.js, custom config), 1C, a test environment, or full control, a VPS is the right choice.

Real-world examples:

• Online store. WooCommerce, Magento, OpenCart already lag on shared.
• WordPress production. With 30+ plugins you hit CloudLinux limits, a VPS gives breathing room.
• 1C / CRM / ERP server. Dedicated environment for in-house systems, terminal access, daily backup, manat billing.
• Automation. Telegram bots, n8n, Zapier alternatives, always-on backend.
• Game server. Minecraft, CS, Rust — local low-latency environment.
• Dev environments. Staging, QA, CI runner, sandbox.
• Database server. Isolated home for PostgreSQL, MySQL, MongoDB.
• Backup hub. A node to collect backups from other servers.

If you get 10,000+ visitors a month, a store with 50+ products, or a corporate application, a VPS is a more correct choice than shared.

VPS vs shared hosting vs dedicated

Shared hosting fits starter projects and small blogs; it is limited for serious workloads. VPS is the ideal balance for medium projects, business apps, growing e-commerce. Dedicated is for genuinely high-traffic sites, large databases, enterprise applications. Most projects start on VPS and move to a stronger plan as workload grows. In our infrastructure, Start → Performance → VDS → Dedicated upgrades are one-click, no complex migration.

VPS in Azerbaijan

When choosing a local VPS, the physical location of the server is the key factor. A VPS in Baku, in a Tier III data center, gives local users low latency, a local IP, and in-country data storage.

Epro.io note — our hardware. Host servers: two Intel Xeon CPUs (Platinum 8268 on the Performance line, Gold 6244 on the VDS line), 2 TB DDR4-3200 ECC RAM, Enterprise NVMe in RAID10, 10 Gbit/s ×2 redundant external uplinks — two cables per server, if one fails the other carries on. We refresh hardware parameters constantly, only upwards: newer CPUs, faster RAM, larger disks. We pick only the fastest memory modules (DDR4-3200) because the gain over DDR4-2400 is up to ~10% in real virtual machine performance. We don’t sell «cheap VPS» on aged hardware. What memory frequency does your current provider use? Which CPU generation? Usually these questions stay unanswered.

Why is a local VPS better than a foreign one?

Latency. A request to a German server takes 60–80 ms, Turkey 30–50 ms, inside Baku <2 ms. The site feels instant to the user.

Local IP (AZ). On an Azerbaijani IP, the site works more smoothly with local SEO, Asan Pay, Million, and local bank APIs.

Compliance. Working with the KGK, tax, government programs? Keep the data in Azerbaijan — legally easier.

Epro.io note — official statuses and licenses. A real business project looks for a certified partner. We have: Tier III certified data center (Delta Telecom DTMDC, Design + Constructed Facility, Uptime Institute audited) · RIPE NCC LIR status and our own AS number (BGP routing, our own IPv4/IPv6 blocks) · cPanel Certified Partner in Azerbaijan (official certified partner status) · State host provider registration · Registration/license for personal-data information resources · 14 years in the market (since 2012), official registration as Epro.io MMC.

Manat billing. No FX fee, e-qaime compatible, simple accounting.

Local support. In Azerbaijani, local team, real people.

Epro.io VPSs sit in the Delta Telecom Tier III data center in Baku. The hardware is ours, no subcontracting. Manat billing, AZ/RU/EN support, 24/7.

FAQ