Hyperconverged Infrastructure

Don’t Waste Your Memory

Guest blog by Mark May, Enterprise Infrastructure Specialist and author of VirtualStorageZone.

In this day and age, businesses make technology purchasing decisions based on many factors. However, more often than not, it comes down to one thing: money. The idea that Hyper-Converged Infrastructures can help organizations save money is nothing new. By buying the entire compute and storage stack together from a single vendor, you should be able to achieve a better price. This is not only common sense, but the premise of warehouse stores: the more you buy from a single source, the better deal you get. And technology is no different, you’re able to save money by purchasing a lot from a single vendor. One thing to consider, however, is how the architecture of a system can impact the total cost of ownership.

One of the most critical elements of any hyperconverged offering is its ability to take local storage and share it out amongst all nodes in the cluster. That is, after all, what makes it hyperconverged. This ability consumes some amount of resources depending on how the system is architected. Architecture of this type fits into two categories: coupled and decoupled.

In a coupled storage architecture, like the one provided by EVO:RAIL, the local storage on the host is presented using the hypervisor kernel. This approach is favorable because it limits the amount of host-based resources consumed by the storage presentation. In a decoupled storage architecture, a virtual machine on the hypervisor host must be created to share the local storage. Running that virtual machine consumes CPU and memory resources. In the pie chart below we see the amount of memory, that is lost using this type of approach.



In the chart above chart, we have a four node configuration.  Each node as 128 Gb of memory, 16 of which is lost to the virtual machine due to the decoupled storage architecture. All in total the cluster is using 64Gb of memory to share storage. A situation where this memory loss is especially important is when using hyperconverged infrastructure for virtual desktops.

One of the challenges when designing a VDI solution is how to keep costs per desktop in line with the non-virtualized alternative.  To achieve this, it is important to be able to run as many virtual desktops on top the cluster as possible. The more desktop instances you run on a cluster, the better the price per desktop will be. Giving up limited CPU and memory means you can fit fewer desktops on a node. If we take a modest virtual desktop with 4gb of memory, we lost out on 16 desktops on the cluster.  The result is that we can only fit 28 hosts per host. Using the EVO:RAIL software defined storage architecture, you could fit 32 desktops per host.  To run the same number of virtual desktops with a decoupled SDS layer, an additional host would need to be used.

If you are wasting CPU and memory resources with an inefficient architecture, you are throwing money away. EVO:RAIL, with its kernel level integrations, allows for the most efficient use of resources. As you research various hyperconverged offerings, it is important to understand the limitations of the various architectures. The EVO:RAIL architecture is designed from the ground up for efficient use of resources. In this manner, the total cost of ownership of the solution can be driven down when compared to other offerings.


MarkMayFor over twenty years, Mark has been deep in the trenches designing enterprise-class data center solutions. Starting out his career in 1995, Mark embraced his entrepreneurial spirit to co-found a small internet service provider in Ohio. During this time, he helped many other organizations, both small and large, to develop network solutions. It was during these years Mark was first exposed to the world of enterprise storage, where he eventually focused his efforts.

Working in the enterprise space has given Mark has an excellent understanding of the challenges facing the modern infrastructure provider. With a deep understanding of storage, networking, and virtualization he can help clients design and implement new solutions to address these challenges. Mark is an energetic speaker who seeks to entertain and educate people. He has written many articles on storage, networking, and virtualization technologies.



Follow us on Twitter at @vmwevorail and our blog, Virtual Blocks, for future updates and to see what EVO:RAIL can do for you.




Leave a Reply

Your email address will not be published.