In any data center, there has for decades been one ... infrastructure in a data center, and yet for a variety of reasons
© 2017 Liqid, Inc.
www.liqid.com
ANALYST REPORT
How Liqid is Dominating the Next Big Thing in the Data Center By: Arik Hesseldahl Analyst & Journalist
© 2017 Liqid, Inc.
www.liqid.com
In any data center, there has for decades
Colorado called Liqid has established
been one fundamental problem that’s hard
itself as an early leader.
to get around: Business needs are fluid yet hardware is not.
Put simply, composable infrastructure makes every part of the IT stack —
Any time that the needs of a business
computing, storage and networking
change, and they change often, IT
and even GPUs — so flexible that it can
environments tend to be incredibly
be reconfigured on the fly as the needs
inflexible. Making changes can quickly
change.
spiral into expensive, time-consuming problems that can slow a business down,
The economic implications by themselves
creating openings for competitors.
are significant. IDC recently estimated that 75 percent of IT budgets are devoted
One reaction to this has been the
to resources that are under-utilized. What
“software defined” movements in both
this means is that in a typical enterprise
networking and storage. Resources that
environment under normal operating
were once limited by the specifications of
conditions, the vast majority of available
their hardware have been opened up by
computing capacity sits idle.
APIs and commodity hardware and made more flexible. But despite the potential,
You wouldn’t want to own a factory using
software defined networking and storage
less than a third of its capacity because
has made a limited impact on data centers
owning equipment that’s doing nothing
overall.
is expensive. The same is true of IT infrastructure in a data center, and yet for
But there’s a new approach called
a variety of reasons, utilization rates of 15
composable infrastructure that has been
percent and less are the norm.
gaining steam lately, and a company in
02
© 2017 Liqid, Inc.
www.liqid.com
Companies like Google, Facebook and
percent. And on weekends it can be set
Amazon, whose data centers are often
to devote a much smaller pool to that
said to boast utilization rates are closer
app, allowing the available hardware to be
to 30 percent and sometimes higher are
put to other uses, improving the overall
the exception to this rule. But they’re not
utilization of the resources.
going to solve this problem for anyone but themselves. Liqid is promising to help
This young company — it has raised
companies reach target utilization rates of
more than $20 million in two rounds of
up to 90 percent.
VC funding — has a handful of proof-ofconcept projects underway with large
How Liqid does this is interesting. It has
companies but has yet to name any
created a PCIe switch that is placed atop
customers. The potential use-cases are
a standard IT rack. That switch allows for
many. Big enterprise institutions which
the dynamic creation of servers configured
operate their own infrastructure could add
at the bare metal level, where there’s no
Liqid switches to boost the efficiency of
virtualization software to add complexity
their applications and platforms, making
or the added expense from licensing fees.
them more competitive.
Once deployed Liqid allows IT resources
Emerging applications like artificial
to be pooled from as much or as little of
intelligence (AI), virtual reality (VR), and
the available hardware as needed in an
3D rendering consume huge amounts of
arbitrary manner. Applications get the
computing resources. Specialized GPU
optimal level of resources they require, no
chips can speed up that process, but until
more, no less. And if needs change, it’s
now there’s been no way to pool them.
easy to re-configure on demand, and even
Liqid makes that possible too, reducing
in an automated way.
a high compute workload that previously required days to one that now requires
Let’s say a workplace application gets
hours.
especially busy around 10 AM on weekday mornings as everyone in the office signs
Overall Liqid is offering its customers
on and gets to work. Liqid allows IT
more flexibility from their existing
managers to set automated policies to
infrastructure than was possible with
spin up more resources — all the compute,
previous solutions. And when they invest
networking and storage needed for a
in new hardware, they can either get more
work day —from the available pool of
computing bang for the buck, or scale
hardware. Liqid enables the automatic
back their budgets and simply not buy
reconfiguration of those resources once
extra gear they don’t need, essentially
demand reaches a pre-set level, say 80
enabling them to do more with less.
03
© 2017 Liqid, Inc.
www.liqid.com
WHY LIQID IS THE LEADER IN COMPOSABLE Liqid also has some important wind at its
Liqid, it turns out, is the only composable
back. Earlier this year semiconductor giant
infrastructure player in the marketplace
Intel released version 2.1 of its Rack Scale
today that can unite all the different
Design (RSD) specification, basically a
elements of the stack — compute, GPU,
guide to what the Intel thinks data center
storage and networking — through this
infrastructure will look like in the near
common PCIe connection.
future. Liqid’s approach to composable is very much aligned with the promise of
It did this by making a PCIe fabric switch
Intel’s RSD.
to interconnect the pools of resources, because the interface enables all
A key plank of that specification includes
the major datacenter elements and
the use of PCI Express as the preferred
peripherals to communicate natively.
interface for connecting all pools of
Liqid’s management console software
resources together. There’s a handful
keeps track of it all.
of reasons that PCIe is a better choice than Ethernet, which is what a competing
Amazon, whose data centers are often
composable infrastructure solution called
said to boast utilization rates are closer to
Synergy from Hewlett Packard Enterprise
30 percent and sometimes higher are the
favors. For one thing Ethernet is slower
exception
and tends to suffer from higher latency than PCIe. Additionally, Ethernet is not a native interface for connecting to GPUs or to SSD storage, which PCIe is.
04
© 2017 Liqid, Inc.
www.liqid.com
DISAGGREGATED VERSUS CONVERGED Another key aspect of Liqid’s approach is
GPUs in particular are the new cool kid
that it enables a disaggregated approach
in the data center, adding some extra
to hardware. This means that instead of
computing muscle to specialized work
a single piece of hardware that combines
that requires more time and energy
compute, memory and storage resources,
with traditional CPUs. Liqid’s ability to
it offers the ability to connect standalone
harness hundreds of them into a single
compute box, storage box, or GPU box to
cluster is unique among the composable
scale out only the required resource.
infrastructure players.
The company believes that the
Down the road, I think the most exciting
disaggregated approach is the right one
direction composable infrastructure may
for one key reason: Flexibility. Customers
take lies with its potential for machine
only have to pay for resources they need
learning and artificial intelligence. Already
and can forego the ones they don’t. This
Liqid’s management tools allow for the
approach differs from what other players
use of policies and rules that trigger
in the composable space offer. Most favor
automated responses under certain
the converged approach of selling boxes
conditions.
that combine all the datacenter elements, whether or not they are necessary to what
Eventually systems will learn from patterns
a given user is trying to achieve.
of usage and demand and configure themselves in ways that can’t help but
What this effectively means is that you can
evoke philosophical discussions that
buy precisely the mix of compute, storage,
include references to science fiction
and GPUs that meets your needs, without
movies, but which will in day-to-day reality
having to pay for additional elements that
save time, energy and the human effort
you might not want or need.
associated with the massive undertaking that managing a data center already is.
05