The link between hyperconvergence and web-scale or “Why Hyperconvergence Needs Web-Scale”

Hyperconverged systems are being discussed with increasing regularity by the trade press and analyst community. VMware’s introduction of EVO:RAIL in August this year validated this new model for infrastructure that Nutanix introduced a few years ago.

Hyperconvergence is a fundamentally different architecture from traditional three-tier infrastructure solutions that are prevalent in data centers today. Hyperconvergence is an architectural model for IT infrastructure in which compute and storage resources are physically packaged in small servers, typically x86 based.

Hyperconverged systems are by their nature distributed systems. The building block of a hyperconverged solution is a physical x86 server with a processor, memory and storage (including Flash). Customers deploy these solutions by buying several of these servers or nodes and connecting them to a top-of-rack Ethernet switch.


Figure 1: Scale-up storage in a traditional three-tier architecture 

The distributed nature of hyperconverged systems requires a fundamental rethink of how data and metadata are stored across nodes and accessed for storage operations. In a traditional scale-up storage system, all the data is stored in drives that sit behind a pair of storage controllers. These controllers are responsible for all I/O, as well as for storage operations such as deduplication, compression and snapshots.

Figure 2: Data in hyperconverged systems 

In contrast to this centralized approach to storage, hyperconverged systems have data spread across several nodes. Any centralized treatment of the metadata or storage operations will severely limit the scalability of the cluster and/or introduce points of failure or weakness in the system.

This is where web-scale architectures come in. Companies such as Facebook, Google and Amazon built large-scale distributed systems that delivered the ability to quickly and predictably add infrastructure resources when needed with zero disruption to the system, and to manage the growing footprint with few admins.

In building these distributed systems these companies pioneered a transformative approach to buying, deploying, managing and scaling infrastructure. This is what web-scale is all about. The principles of web-scale are tried and tested rules for building distributed systems the right way.


What does this mean for enterprise IT customers? Not all hyperconverged systems are created equal. If you’re considering hyperconverged systems to simplify your data center and scale your infrastructure predictably on the go, make sure that the solution is built using core web-scale principles and architectures. Some questions to ask:

  1. Can you start with a few nodes and grow in small increments (one node) when needed?
  2. Is data distributed across all nodes in the cluster to ensure no hotspots?
  3. Do all nodes participate in serving I/O without one node limiting throughput or becoming a bottleneck?
  4. Is metadata distributed across all nodes in the cluster for linear scalability without limits, or are there special metadata or name nodes that limit deployment size or eventually degrade performance?
  5. Do all nodes participate equally in storage activities such as deduplication and compression across the cluster to maximize performance by using all CPU resources across the system?
  6. Is the cluster able to withstand failure of individual nodes or components and ensure data availability without data loss? This is an essential foundation for always-on operation
  7. If a node were to fail, do all other nodes participate in the recovery process to accelerate rebuilding of data and quickly restore state.

For Nutanix, you can learn more about our system here: Feel free to reach out to us if you want to talk about our technology further.