Welcome to the fourth and final part of our series of posts on The five dangers of Virtual Servers.
Danger 4: The danger of over-contention
There’s a potential for over-selling with virtual servers, as the provider may not tell you how many virtual machines they intend running on one physical node, whether memory and CPU time are contended, or how fast the physical node is uplinked to the network.
The whole point is to put limitations in-place to stop one virtual server from hogging the resources at a performance cost to neighbouring virtual servers.
A responsible provider uses a virtualisation technology that does not allow memory or disk space to be contended, and has fixed parameters in place to ensure CPU and network resources are fairly shared out. This includes setting an upper limit to the number of virtual machines that can run on a physical node, and also ensuring the physical nodes are uplinked to the ‘net at a suitable speed such that every server gets a decent sized connection. For example a gigabit connection shared between 30 virtual machines gives an average throughput capacity of approximately 30Mbps to each virtual machine, whereas a 10Mbps connection shared by the same 30 virtual machines would give a measly 0.3Mbps average to each VM.
5. The danger of not having tools to help yourself in the case of emergency.
You would generally want KVM over IP and remote reboot facilities on a dedicated server, to give you “sat in front of the machine” access in the case of a major OS failure, or to correct network settings when the machine is otherwise inaccessible.
The same should apply to a virtual server. This technology can be a life-saver when you need to work on a virtual server at 2am without waiting for your provider to respond to help you out. It gives you complete self-sufficiency.
Conclusion
There’s plenty of scope for virtual server providers to cut corners. That said, if you research your provider well, using the above questions as part of your decision-making process, you’re likely to find a service that’s a high-availability and high-performance alternative to dedicated servers.
To sum up make sure that any virtual server provider can meet the following criteria and you’ve done your utmost to mitigate the dangers we’ve described:
- Uses SAN Storage
- SAN and host servers have redundant critical components
- Nodes dual-uplinked through two switches (front-end and backend)
- You have your own VLAN
- Provides a hardware firewall
- No kernel sharing between host node and virtual machines
- Has a reasonable SLA
- Console access with reboot facility
- Has support that is responsive and knowledgeable
- Knows how to look after customers
As you’ll have no doubt guessed, Melbourne’s UltraVM™ Cloud Servers come up trumps on all of these points.
Daniel Keighron-Foster, Managing Director