The "Burst RAM" Lie: Why Your VPS Feels Slow
It’s 3:00 AM. Your Nagios pager goes off. The load average on your database server just spiked to 25.0, but your traffic is flat. You check top and see plenty of free CPU. You check RAM, and it looks fine. What’s happening?
If you are hosting on legacy platforms like Virtuozzo or OpenVZ, you are likely the victim of a "noisy neighbor." In the crowded budget hosting market, providers love to oversell resources. They sell you "Burst RAM"—memory that doesn't actually exist unless no one else is using it. When the node gets busy, your MySQL queries hit a brick wall.
For serious systems architects, this is unacceptable. The solution isn't just a bigger server; it's better architecture. Enter KVM (Kernel-based Virtual Machine).
KVM vs. The Legacy World
Unlike OpenVZ, which shares a single kernel across all users (a glorified chroot), KVM turns the Linux kernel itself into a hypervisor. This was merged into the mainline Linux kernel in version 2.6.20 (Feb 2007), but it is only now, in 2009, maturing into the killer app for enterprise hosting.
With KVM, your operating system is fully isolated. You have your own kernel. You can load your own modules. Most importantly, memory is hard-allocated. If you buy 4GB of RAM on a CoolVDS KVM instance, that RAM is reserved for you at the hardware level. No more "Burst" marketing tricks.
The Performance Gap
Many sysadmins worry about the overhead of full virtualization. In the days of QEMU software emulation, that was valid. But with modern hardware support found in the new Intel Nehalem (Xeon 5500) processors, the overhead is negligible. Intel VT-x extensions allow the CPU to handle the virtualization lifting.
To get true performance, however, you must use VirtIO drivers. These paravirtualized drivers allow your VM to talk directly to the hypervisor without emulating an ancient IDE disk controller.
War Story: The Magento Meltdown
Last month, we migrated a client running a heavy Magento e-commerce store. They were hosted on a generic "Cloud" VPS in Germany. Every time they ran a catalog re-index, the site timed out.
We ran iostat -x 1 and saw %iowait hitting 40-50%, yet the disk throughput was only 2 MB/s. The physical disk on the host node was thrashing because fifty other customers were fighting for the same spindle.
We moved them to a CoolVDS KVM instance backed by 15k RPM SAS RAID-10 arrays. The result? Re-indexing time dropped from 45 minutes to 4 minutes.
Pro Tip: Tuning Your I/O Scheduler
If you are running on our high-performance storage, the default Linux I/O scheduler (CFQ) might actually slow you down. CFQ tries to be fair, but on a fast virtualized disk, you just want raw speed.
We recommend switching to the deadline or noop scheduler. Here is how you do it on CentOS 5:
# Edit /boot/grub/menu.lst
# Add 'elevator=deadline' to your kernel line
kernel /vmlinuz-2.6.18-128.el5 ro root=/dev/VolGroup00/LogVol00 elevator=deadline
This simple change prevents the kernel from re-ordering requests unnecessarily, shaving precious milliseconds off your latency.
Data Sovereignty in Norway
Performance isn't just about speed; it's about reliability and law. With the current scrutiny from the Datatilsynet (Norwegian Data Inspectorate) regarding the Personal Data Act (Personopplysningsloven), knowing exactly where your data lives is critical.
Hosting outside of Norway introduces latency and legal grey areas. CoolVDS servers are located in Oslo, directly connected to the NIX (Norwegian Internet Exchange). This ensures your data stays within Norwegian jurisdiction and your ping times to local users remain in the single digits.
Architect's Note: Never assume a VPS is backed by RAID. Always ask your provider. At CoolVDS, we refuse to deploy single-drive nodes. It’s RAID-10 or nothing.
Conclusion
If you are running a static HTML site, OpenVZ is fine. But if you are managing a database, a mail server, or a Java application, you need the isolation and dedicated resources of KVM.
Don't let your infrastructure be the bottleneck. Deploy a KVM instance on CoolVDS today and experience the stability of the 2.6.30 kernel on true enterprise hardware.