The "Burst RAM" Lie
It is 3:00 AM. Your pager goes off. Your MySQL slave has stalled. You log in via SSH, and it takes 15 seconds just to get a prompt. You check top. Your load is low. Your RAM is free. So why is the server crawling?
If you are hosting on a budget VPS using OpenVZ or Virtuozzo, you are likely the victim of a "noisy neighbor."
In the hosting market right now, overselling is the standard business model. Providers pile hundreds of containers onto a single host node, banking on the fact that not everyone will use their resources at once. They sell you "Burst RAM"—a resource that exists on paper but vanishes when you actually need it. For a dev box? Fine. For a production Magento store or a high-traffic forum? It is suicide.
At CoolVDS, we grew tired of debugging vague performance issues caused by other people's processes. That is why we standardized on KVM (Kernel-based Virtual Machine).
KVM: Real Hardware Virtualization
Unlike containers, KVM requires hardware virtualization extensions (Intel VT-x or AMD-V). When you provision a KVM instance, you aren't just getting a directory `chroot`'d on a host kernel. You are getting a dedicated slice of the CPU and RAM that the host kernel cannot easily steal back.
If you have 4GB of RAM on a CoolVDS KVM slice, that physical RAM is allocated to your PID. It’s yours. If the guy next door decides to compile the entire Gentoo repository, your database doesn't feel it.
Validating Your Environment
Don't take a provider's word for it. Check if your CPU actually supports the necessary extensions. On your current server, run:
grep -E 'svm|vmx' /proc/cpuinfoIf you see output, you have hardware support. If you are inside a KVM guest, you act like a real server. You can load your own kernel modules. You can tune your own TCP stack. You can even run a custom kernel if RHEL/CentOS 6 default doesn't suit your needs.
The I/O Bottleneck: HDD vs. SSD
The other killer in 2011 is disk I/O. With the flooding in Thailand driving hard drive prices up and availability down, many hosts are squeezing old 7.2k SATA drives past their limit. In a virtualized environment, random I/O performance (IOPS) is king.
We have begun rolling out RAID 10 SSD setups. The difference is not subtle. While a standard SAS array might give you 300 IOPS, a solid-state array pushes thousands. For database-heavy applications, this reduces query latency from "noticeable" to "instant."
Pro Tip: Tuning the Scheduler
If you are running on our KVM SSD storage, the standard Linux CFQ scheduler is often overkill. It tries to reorder requests for spinning platters that don't exist. Switch to the `noop` or `deadline` scheduler to lower CPU overhead.
Edit/boot/grub/grub.confand append to your kernel line:elevator=noop
Data Sovereignty in Norway
Latency is physics. If your customers are in Oslo, Bergen, or Trondheim, hosting in a datacenter in Texas makes no sense. You are adding 120ms of lag before the first byte is even sent.
But beyond latency, there is the legal aspect. With the Personopplysningsloven (Personal Data Act), Norwegian businesses have a responsibility to secure customer data. Hosting locally at NIX (Norwegian Internet Exchange) connected facilities ensures your data stays within jurisdiction and under strict Norwegian privacy standards. Datatilsynet (The Data Protection Authority) is becoming increasingly vigilant about where sensitive data lives.
The CoolVDS Architecture
We don't believe in magic. We believe in engineering. Our stack is simple:
- Hypervisor: KVM on CentOS 6
- Storage: Hardware RAID 10 (Redundancy + Speed)
- Network: Gigabit uplinks directly to NIX
- Management: Libvirt & Custom Panel (No bloat)
If you are tired of fighting with `User beancounters` limits on OpenVZ or wondering why your ping times fluctuate wildly during peak hours, stop sharing your kernel.
Get a dedicated environment that behaves like metal.
Deploy a test KVM instance on CoolVDS today. We can have you root on a fresh CentOS or Debian Squeeze install in under 60 seconds.