|
Infrastructure
|
Our
virtual servers
run on our sophisticated server farm, located in a carrier-neutral
premier-class data centre near London in the UK. We monitor, service
and upgrade the underlying infrastructure, providing you with high
availability and the capacity to scale up rapidly.
Our virtual servers offer all the power and control of a
traditional hosted server, whilst giving you the benefit of:
-
Instant flexibility: to scale your machine up or down using any web browser
-
Peace of mind: of running on professionally managed infrastructure, with
automatic failover and regular automatic snapshots of your entire
machine state
-
Cost efficiency: of buying exactly the capacity you need on an infrastructure built
to scale, and growing only when you need to
-
Personal support: expert specialist technical support with a friendly face
The technical details
Each layer of our infrastructure has been carefully designed and
configured by our team to provide optimum performance to your virtual
servers. We have deep understanding of this stack, and have
contributed to the development of several of the software packages
which we use in order to add features which we needed to provide you
with our flexible, safe and efficient service.
-
Virtual server environment:
you have complete control to choose the operating system,
applications and configuration of your machine, just like a physical
hosted server. In addition, you can use any web browser to scale
your machine up or down and to manage all of the
full suite of services
that you need for your web application (e.g. domain registration,
DNS).
-
Virtualization platform:
we use
Linux KVM, the advanced virtualization platform built into mainline Linux.
As described below, we believe KVM is superior to Xen, VMWare and
Virtuozzo.
-
Clustering technology:
we cluster our servers for redundancy, load balancing and automatic
failover, using
RAID 6
to keep your data safe and your virtual server running even in the
(unlikely!) case of either an entire disk server or two separate
hard disks failing at once (one better than mirrored solutions or
RAID 5).
-
Server OS:
we run our own stripped-down Linux distribution on our servers to
host the virtualization platform with maximum efficiency and
security.
-
Server hardware:
we use high-end multi-processor servers, fitted with quad-core AMD
"Barcelona" Opteron CPUs (B3 revision). These are the first
processor released by either AMD or Intel to offer
second generation hardware support
for virtualization, greatly increasing performance and effectively
eliminating MMU virtualization overhead.
-
Network connnectivity:
we connect via Connexions4London's
Multihomed IP Transit, which gives combined multiple connectivity into Tiscali, NTT and
Level 3's Tier 1 networks with peering to over 400 ISPs in the UK,
Europe and US, via all the major exchanges, such as LINX and LONAP.
-
Data centre:
our server farm is located in
BlueSquare 2, a carrier-neutral premier-class data centre in Maidenhead, UK
(near London).
Why we chose KVM virtualization
Before selecting
Linux KVM, we evaluated VMWare and Xen, two other
virtualization platforms providing similar capabilities to KVM (Virtuozzo
only provides OS-level containers). We selected KVM as the best
architecture for virtualization on modern processors with fast
hardware virtualization support (VT-x and NPT on Intel or AMD-V and
EPT on AMD).
Increasing hardware virtualization support
- Historically, virtualization platforms used software to trap and
simulate certain instructions, memory management and I/O in the
host virtual machines. VMWare was an early leader in this
software technology.
- With the first generation of hardware virtualization, the
VT-x/AMD-V extensions trapped these instructions in hardware,
giving a significant speed improvement. However, virtualized
memory management and I/O remained bottlenecks. Xen was an early
proponent of paravirtualization, which attacks those bottlenecks
by modifying the host operating system at compile time.
- With the second generation of hardware virtualization, the
NPT/EPT extensions minimize the memory management bottleneck.
As a result,
MMU paravirtualization is a legacy approach, leaving just scheduling and I/O to be virtualized in software
by a hypervisor. (I/O virtualization requires a good set of
device drivers for the underlying hardware, of course: an area in
which Linux excels.)
Hypervisor architecture and device drivers
-
Linux KVM
is a hypervisor which is built into mainline Linux. It
uses the full range of hardware virtualization support, and
directly uses the regular Linux scheduler and I/O device drivers.
- Xen runs an
external hypervisor
for scheduling, and uses a modified Linux kernel in domain 0 to
provide device drivers.
- VMWare runs a proprietary external hypervisor, which includes
scheduling and
device drivers, many of which are adapted from Linux.
- We believe the KVM architecture is superior to both Xen and
VMWare, since the
mainline Linux
scheduler and device drivers are both extremely well designed,
widely deployed, professionally maintained and throughly tested, to a
level likely well above what a single company can achieve on
either their own proprietary codebase or locally maintained fork
of Linux.