I remember the day I saw my first vMotion – unbelievable it has been that long ago – but now VM live migration is a standard for all hypervisors. When we all decided to adopt virtualization we had some clear benefits in mind, and those boil down fundamentally to better resource utilization, reduced datacenter footprint, energy-saving, faster server provisioning, increased uptime, improved disaster recovery, isolated applications, extend the life of older applications and workload portability.
We never thought of virtualization as an enabler for performance enhancement because we knew that workloads would be sharing hardware resources such as processors and network interfaces; but the virtualization benefits easily outweighed the notion of having dedicated servers with low utilization; and as long the applications received the resources required to work properly we would consider that a fitting result.
Twelve years later virtualization is standard and the public cloud is similarly becoming a first class citizen for enterprises. Today, when we think about public cloud we don’t think in terms of bottoms up performance, but rather scalability, implementation speed and flexibility. The large majority of public cloud providers don’t even guarantee peak performance, but as long as applications receive the correct amount of resources to operate efficiently then application owners are pleased with the results.
Hyperconvergence on the other hand has the capability to bring together the best of both worlds; combining the benefits of virtualization and public cloud deployments is what is often referred as private cloud. The economics and technology for creating private cloud deployments only start making sense with hyperconvergence, where buying, deploying, managing and scaling datacenter infrastructures become less expensive and time-consuming.
That brings me to the hyperconverged storage performance discussion that I have previously discussed in “Is Hyper-Converged faster than SAN?”. Does optimal storage performance in a hyperconverged platform matter without considering the benefits associated to the solution?
The answer is No, because this is not a drag-race but rather a datacenter transformation technology, and as long as applications are efficiently delivered the benefits will always outweigh the premise that having the fastest performance is best. Josh Odgers has a very good article on Peak Performance vs Real World Performance that illustrate why performance is rarely a deciding factor for a storage solution.
With the Nutanix platform the benefits are numerous and they define a baseline as to what to expect from a hyperconverged platform.
- Granular (node-by-node) and linear scalability with pay-as-you-grow model.
- Distributed architecture built to tolerate multiple simultaneous failures and heal itself when a failure occurs.
- Combination of storage and compute performance that deliver on business and application needs.
- Storage capacity savings mechanisms to reduce storage consumption.
- Integrated replication and disaster recovery technologies.
- Non-disruptive upgrades for the entire stack.
- Readily available API’s to enable DevOps and automation workflows for the entire datacenter stack (compute, storage and network).
From a features perspective Nutanix offers: de-duplication, compression and erasure coding for capacity savings; synchronous and asynchronous replication with ability to move workloads between sites; backup to public cloud (Azure and QWS); multi-hypervisor support, one-click upgrades etc.
In my personal opinion the discussion about storage or disk performance is as irrelevant as the initial virtualization performance discussions in the early 2000’s because as long as applications are efficiently and cost-effectively delivered with the appropriate SLA’s, the benefits will always outweigh the premise that having the fastest disk performance is best.
Hyperconverged solutions work the whole body at once and must be examined with these context and principles in mind.
If you want the fastest attainable performance, just stick few standalone PCIe storage cards or some memory channel storage inside the server and call it a day. However, if you want to truly transform the way you do IT then there are better options. With the momentum around hyperconverged in the market, my guess is that this will be looked back on as a fundamental technology shift in years to come.
This article was first published by Andre Leibovici (@andreleibovici) at myvirtualcloud.net