A hyper-converged appliance offers a low-risk option for building a cloud computing infrastructure. These preintegrated...
systems include a commercial, off-the-shelf server and multiple drives, as well as requisite software that allows those drives to be shared across a compute cluster.
For any company with limited hardware expertise or IT staff, hyper-converged infrastructure can offer simplicity and speed -- though potentially at a premium price. Still, many private cloud deployments are in their early days, so a pilot investment in a hyper-converged appliance could be a good way to get your feet wet with the technology.
Why does hyper-convergence make sense for private cloud deployments?
The storage industry is shifting away from the old RAID array concept towards small appliances with just 12 drives being the standard for a 2U server. This fits the requirements for many private clouds, as well, since this box size will hold a dual CPU x86 or ARM server and a bunch of dynamic random access memory.
You don't need a lot of CPU power to drive a networked storage appliance, so the other cores are available for virtual machines to run applications. This is the derivation of hyper-converged infrastructure; the storage pool essentially takes over the compute function, as well, which is one reason storage companies are prominent in the hyper-converged appliance space.
From a performance perspective, hyper-converged infrastructure effectively matches system and network bandwidth to the performance of midtier solid-state drives (SSDs). A couple of 10 gigabit Ethernet (GbE) links provide enough network bandwidth, and dual CPUs provide enough compute, to use the SSDs at or near their limits, even when shared across the cluster. However, the typical, off-the-shelf hyper-converged appliance may lack adequate network bandwidth unless it supports 40 GbE with remote direct memory access across the cluster.
Due to the use of virtualization, hyper-converged infrastructure is much closer to a software-defined infrastructure than any traditional server storage model. Full integration and orchestration, with a link to virtual switches presented by software-defined networking, will allow IT teams to control storage administration through code. At the same time, a policy-driven management system allows private cloud tenants to control virtual resources without having to resort to central IT admins for virtual LAN setup and storage allocation.
Network virtualization in hyper-converged infrastructure
In hyper-converged infrastructure systems, networks are also virtualized. Today, the switch function is still embodied by a discreet switch module, but virtualizing network functions frees the switch from a lot of proprietary code. These bare-bones switches are inexpensive, but still deliver on bandwidth.
HCI, coupled with good orchestration software, simplifies private cloud storage allocation, as well, since all the server drives form a pool within a single namespace. The policy approach simplifies private cloud control, as well as the usage billing that should be a part of any cloud deployment. Over time, HCI should allow cloud admins to differentiate service levels for a broad variety of instance types, with allocated bandwidth and IOPS based on the instance.
The future of hyper-converged infrastructure
Hyper-converged infrastructure is building market share rapidly, primarily due to its ease of use. Looking to the future, most pundits believe hyper-converged appliances will take over the IT infrastructure market. This seems likely, but technical innovations may further shake up the market.
The most likely change is the removal of drives from the hyper-converged appliance itself. Instead, we'll see storage-class memory (SCM) replace drives as primary storage. This type of memory will talk directly to the LAN and be shared just like SSDs in current hyper-converged infrastructure implementations, as well as be on an ultrafast memory bus within the local system.
In this model, servers shrink to a much smaller footprint, since SCM is low-power, compact solid-state storage and not intended for hot swapping. At the same time, Ethernet-based storage will create a bulk secondary storage tier for cold data.
Hyper-converged infrastructure is another big step in the commoditization of IT platforms. As the vendor base expands in the next two years, we can expect competitive pricing and a wide variety of alternatives, including a graphics processing unit-based hyper-converged appliance for big data analytics and scientific computing.
Explore hyper-converged architecture for storage
Find out what features you need for a hyper-converged architecture
Add more storage capacity to a hyper-converged appliance