The EuroVPS blog
Field notes.
From the engine room.
Decisions, investigations and lessons from running production infrastructure. Written by the people who look after it.
The archive
More from the team.
A real VPC is a property of the hardware, not a row in someone else's database.
Per-tenant isolation on a leaf-spine fabric, multi-homed resilience, and an encrypted mesh between datacenters: what a private network means when it is not a checkbox in someone else's console.
How we change routing without asking permission.
When most hosts need a routing change, they open a ticket and wait. We make the change ourselves, under a written runbook that keeps it safe.
What the word SSD hides about your host's storage
The three architectures hiding behind the word SSD, what a dedicated all-flash array actually buys you, and the five questions that get an honest storage answer out of any host.
Why your VM has no swap
The VMs we build ship without swap. Paging to flash makes a memory shortfall slower to see and more expensive to serve, and the fix is the one it has always been: RAM.
The meter on your disk, and the array without one
On a rented volume, provisioned IOPS are billed by allocation. Here is what the same disk looks like when the all-flash array underneath it is yours.
How we prove the door is shut before go-live.
Before a site goes live, we attack the true origin from a box our own firewall was never told to trust, scan all 65,535 ports, and confirm the door is shut the way an attacker would see it: connection refused, not a login page.
Why your new server benchmarks slower than the old one when it isn't.
A common question after a migration is whether a lower bogomips reading means a slower server. It does not: bogomips measures nothing you would want to buy. Here is how to prove it with the real work.
How we prepare a news site for a spike.
Traffic finds whatever ceiling the configuration sets. Preparing a news site for its best day is mostly arithmetic and reading the right numbers, not buying hardware. Here is the method.
Redundant, according to whom?
"Fully redundant network" is a checkbox that means nothing unless the party claiming it controls the failover, and a tenant on someone else's backbone is redundant only by permission. Unlike an uptime number, this one you can verify yourself on the public routing table. Here is how to read it, for the host you have now and the one you are weighing.
Your DR traffic commutes on the public internet.
Every compliance review asks where your data lives. Almost none ask what it rides between sites: replication, DR sync, failover traffic. Some providers send replication across the public internet. Ours rides an always-on IPsec mesh between our datacenters, by default, because we own the network at both ends.
Why your email lands in spam (and what it says about your host).
Receiving servers score your sending IP, its reverse DNS, and the reputation of its neighbours before they read a word of your content. Half of that score is inherited from your host. Here is the half you control, in the right order, and the half you should be checking before you sign up anywhere.
We built a VPN Control Center for an agency's own team.
A multi-brand agency asked for self-service: make our own people able to run routine VPN changes. What we built, why it took the team that operates the infrastructure to build it, and what we kept.
A knowledge assistant that never leaves your building.
The most-wanted enterprise AI feature is an assistant that answers from your own contracts, tickets, and wikis. The blocker is always the same: those documents cannot go to a third-party API. They don't have to. The whole pipeline runs on one single-tenant GPU server, and nothing leaves it, including the part everyone forgets: the vector database that is effectively your company, encoded.
A 70B model really does fit on a 48GB card, once you count what quantization costs.
Two teams fail at GPU sizing in opposite directions: one balks at the 80GB card and drops the project; the other crams everything into the cheapest GPU and loses quality it never measured. Same missing conversation. The VRAM math anyone can check, what each card class honestly serves, and what quantization actually costs, shown so you decide.
Your AI vendor rate-limits your best day.
Every builder on a hosted AI API has met the trio: the 429, the capacity-unavailable error, the latency that stretches at peak. They cluster on your biggest days because that is when everyone's biggest days collide. That is not bad luck. That is the seating chart, and you don't control your seat.
Leaving the API is one line of code.
Teams stay on hosted AI APIs long after they want out, because someone said the sentence that ends the meeting: we're too integrated, it's a rewrite. For the plumbing, the rewrite does not exist. The serving world standardized on one wire format, and the private endpoint we hand over speaks it natively. Your SDK doesn't know it moved.
Your AI endpoint doesn't have to be on the internet.
The standard architecture diagram puts the model behind a public URL and hangs a padlock icon on it. Nobody asks the prior question: why is the endpoint on the internet at all? Your documents are private, your app is private, the machine is yours alone. The public leg exists by default, not by need, and optional exposure should be declined.
Stop comparing GPUs by the hour.
The comparison sheet has six providers and one column, dollars per GPU-hour, sorted ascending, decision gravitating to the top row. Nobody asks what the number is a price of. It could be a consumer card, a shared machine, or dedicated datacenter silicon in an EU rack, and the column cannot tell you which.
What a virtual data center (vDC) is, and when it is the right choice
A virtual data center gives one organisation its own pool of compute, memory, storage and network, managed as a unit, with the freedom to create what it needs inside it. This is what a vDC is, how it compares with the alternatives, and how we deliver one.
Configuring WordPress Multisite with Nginx and PHP-FPM on RHEL 9 and 10
A complete, working setup for WordPress Multisite on nginx and PHP-FPM on Red Hat Enterprise Linux 9 and 10 and compatible systems: packages, PHP-FPM pool, nginx server blocks for subdirectory and subdomain networks, TLS, SELinux, firewalld, wp-config.php and verification.
Trusted Hosting Partner with ISO Certifications
EuroVPS is certified to ISO/IEC 27001 (with ISO/IEC 27017 and 27018), ISO 9001, ISO 14001 and ISO 37001. Here is what each certification covers.
Most read
The guides readers come back to- 01Top 10 Reasons Your Site Loads Slowly (And How to Fix Them)
- 02Important Linux Log Files You Must Be Monitoring on RHEL 9 and 10
- 03Configuring WordPress Multisite with Nginx and PHP-FPM on RHEL 9 and 10
- 04Archimedes and the compound pulley: how one man moved a ship
- 05Ultimate Guide to Understanding HTTP Status Codes: 300 Series
- 06Ultimate Guide to Understanding HTTP Status Codes: 400 Series
- 07How much RAM, CPU, Storage, and IOPS resources does your VPS really need?
Put the reading to work
Have a workload in mind?
Start with the numbers.
The sizing desks turn your requirements into a proposed configuration and an indicative price range. An engineer can help you review what fits.
