What Hardware Runs a Cloud Server? Inside a Data Centre

The Physical Reality Behind "Cloud"

The word "cloud" suggests something intangible — data floating in digital space. The physical reality is a rack-mounted enterprise server in a data centre: two or four processor sockets with 16–64 cores each, 256–1,024 GB of ECC RAM, NVMe SSD arrays delivering millions of IOPS, and 25–100 Gbps network connections. When your Tally company database opens in 3 seconds instead of 25, that speed comes from these specific hardware components.

The CPU: Many Cores, High Clock Speed

Enterprise servers use processors like Intel Xeon Scalable or AMD EPYC — CPUs designed for server workloads rather than desktop use. Key characteristics: many cores (16–64 per socket), ECC memory support, and high sustained performance under continuous load.

For virtual machine (VPS) hosting, the physical server's cores are divided among virtual machines. Your VPS receives dedicated vCores — real CPU cores allocated exclusively to your VM, not shared with other tenants. When M A Global Network specifies "4 dedicated vCores" for a 5-user Tally setup, those 4 cores are reserved from the physical server's pool exclusively for your VM.

RAM: ECC Memory at Enterprise Scale

Enterprise servers use Error-Correcting Code (ECC) RAM. ECC memory automatically detects and corrects single-bit memory errors — the type of low-level hardware fault that can cause silent data corruption on non-ECC systems. For financial databases running Tally or Busy, ECC RAM provides an additional layer of data integrity that consumer-grade hardware cannot match.

Physical servers hosting cloud VPS instances typically carry 256–1,024 GB of RAM, divided among VMs. Your VPS receives dedicated RAM allocation — not drawn from a shared pool. Each user session in Tally requires approximately 2 GB of the server's RAM — dedicated, never contended.

Storage: NVMe SSD Arrays

The storage subsystem is where the most dramatic performance differences between cloud and local office hardware appear. Enterprise NVMe SSD delivers 5,000–7,000 MB/s sequential throughput and 300,000–700,000 random IOPS. A typical office server's SATA SSD delivers 500 MB/s and 50,000–80,000 IOPS — 6–10× slower for the random small-block operations that dominate Tally and Busy database workloads.

Data centres use NVMe SSD in RAID configurations — typically RAID-10 or RAID-6 — providing both performance and redundancy. A single drive failure in a RAID array does not cause data loss or service interruption. The array continues operating on remaining drives while the failed drive is replaced.

Network: Enterprise Fabric at 25–100 Gbps

The network inside a data centre operates at 25–100 Gbps — hundreds of times faster than typical business internet connections. This internal network speed is relevant for backup transfers, storage I/O, and communication between components. The connection from the data centre to the internet (the "uplink") is typically 1–10 Gbps with redundant providers — no single ISP failure takes the server offline.

For Indian users accessing cloud servers in Indian data centres, latency is 10–30 ms — fast enough that Remote Desktop sessions feel indistinguishable from local applications to most users.

Virtualisation: How One Server Becomes Many

A hypervisor (virtualisation software such as VMware ESXi, Microsoft Hyper-V, or KVM) runs on the physical server and divides its resources among multiple virtual machines. Each VM has an isolated hardware environment: dedicated CPU cores, dedicated RAM, and dedicated storage allocation. VMs cannot access each other's resources or data — complete isolation at the hardware virtualisation layer.

This isolation is what makes cloud VPS both secure (your VM is isolated from others) and reliable (another customer's VM having issues does not affect yours).

What This Means for Your Tally or Busy Performance

Frequently Asked Questions