What’s the Difference Between Bare Metal and VM Servers?

10 Jul 2023
Nannette Vilushis
Product

Bare metal servers are dedicated physical machine resources offered by a cloud service provider. The entire machine is dedicated to the ISV. The operating system is loaded directly onto the machine, the ISV’s app runs on that OS, and the app has exclusive access to all the computing resources on that machine.

In contrast, virtual machine (VM) servers run on a physical machine but require a hypervisor to be installed on the machine in addition to the operating system.The hypervisor partitions the physical machine into multiple VMs, each running its own operating system. While each VM runs independently, all VMs on a physical machine share that machine’s computing resources.

When an ISV delivers its app to users from a VM server, the hypervisors required to partition the physical machine into VMs consume 5 to 10% of the server’s resources, resulting in a very slight latency. For most ISVs, this latency makes a negligible impact on the user experience.

But—if the ISV is also using a virtual desktop solution like Citrix® to deliver its application, latency is going to be more problematic. Why?

Citrix complexity creates latency.

Here are just a few examples:

Out of the box, Windows is not optimized for Citrix—experts count 100 or more possible optimizations that can be done to reduce latency.
In addition to installation and setup of Citrix itself, implementing a Citrix environment requires coordination of many subsystems, including SQL, storage, and Microsoft® Active Directory. Any miscoordination affects latency.
Issues in the overall design, even small ones, can add latency; for example, a VM that’s even slightly undersized.
Misconfigured user profiles will adversely impact application performance.
Citrix Gateway physical placement—to minimize latency, the Gateway should be placed close to the server.
Citrix NetScaler can slow network traffic due to misconfiguration, or issues out of IT’s control, like CPU or memory spikes and retransmission timeouts.