Hyperconverged infrastructure (HCI) became main stream with
Cisco recently introducing a new suite of products for this segment. It is
particularly interesting in view of the fact that the traditional scale-up solution vendors have been dismissing this
technology and pushing against this architecture with “converged” mash up of
traditional compute, storage and switching products conveniently prepackaged
and loaded with software.
Certainly
the converged solutions solve some pain points when deploying the“traditional
IT” solutions but proponents of hyperconvergence would argue that it is time to take another approach. It’s
time to consider parallel architectures similar to those used by the large hyperscale businesses. HCI combines
compute and storage tiers into smaller bite sized pieces that are stitched together
with software to form an overall pool of resources. Need more capacity? Just
add more nodes and they automatically become part of the overall resource pool.
Over the
past few years this scale-out HCI
architecture has been pioneered by a group of start-up companies such as
Nutanix and Simplivity. They have suggested that this architecture sprang from
concepts developed for hyperscale
networks and carries over a distributed element theme where hardware
resources are more generic and software becomes the dominant technology.
It
claims to provide the advantages of fault-tolerance, agility and scale built
again on the hyperscale design models. The software is quite different however
in that it supports a variety of virtual machines popular with enterprise
customers, VMware, Hyper-V, KVM. Some hyperscale vendors also offer their own
virtual machine platforms that can be attractive alternatives given the heavy
licensing costs involved with some of the larger VM players. From the
administrative perspective, each new hardware resource is automatically added
to the resource pool. The automated administration tool set is also a concept
that they say mimics the ease and efficiency of hyperscale designs
Some early adopters have shared some success
stories with me. Others have been reluctant to adopt a new and perhaps more
risky approach but could see some value to their virtual desktop infrastructure
programs for example and would be launching trials to prove in the business
case. Now it seems that the question of a fringe technology will no longer be
an issue as HCI is now a serious contender for cloud agile network services
endorsed by a leader in the networking market.
Implementing
HCI has been primarily a discussion regarding Ethernet architectures/topologies
and performance. The distributed processing and storage hardware obviates the
need for dedicated SAN infrastructure. The capacity of HCI increases as more
and more of these distributed elements are added to the network. They are all
abstracted into one large pool of compute and storage resources that service
workloads as required. The software control layer takes care of resource
allocation and duplicates files and processes to mitigate risk of hardware
failure. The resource pool can be distributed across multiple racks, multiple
rows, halls or even over multiple data centers similar to the approach taken by
hyperscale systems.
Ethernet fabric networks seem to be
ideally suited to enable HCI. The strong support for east-west traffic, fault
tolerance and scale match the design paradigm of HCI. The
ability to set low latency and low contention levels is critical in allowing
fabrics to tune the designs to track the growth of this architecture.
Using a physical cabling infrastructure enabled with connection point identification (CPID) technology supports
fabric automation at all levels of the stack, by providing real-time data about
the health of the physical infrastructure, and its suitability to transport
high speed data across any link or channel in this complex mesh of cables
between the leaf and spine switches. Fabric network design requests are on the
rise – due to HCI and other cloud like architectures that are becoming the new
normal in enterprise IT.
CommScope is a world leading provider of pre-engineered
and pre-manufactured network infrastructure solutions, that like HCI, are
modular in construction, easy to deploy, and easy to design and administer. Examples of these solutions
include the Rapid Fiber panel and InstaPATCH360 pre-terminated
fiber solutions.
Additionally CommScope’s Quareo product set, enabled with CPID technology, gives network administrators an unparalleled real time insight into
the physical layer, in a way not previously thought possible. All these
CommScope technologies are prioritized to support fabric speeds and
applications much like HCI is prioritized to support cloud based computing
resources.
Is HCI a
giant step for enterprise networks? In
my opinion it certainly seems to help move many legacy applications in the
direction of a cloud native environment. If HCI makes sense, are fabric-based Ethernet networks in your
future as well? I would love to hear you feedback.
Also,
come to my presentation at the upcoming Datacenter
Dynamics Enterprise show in New York on April 19, 20156 at 3:30 p.m. EDT.
What else would you like to know about hyperconverged
infrastructure?