With enterprises clamoring for hardware and software that helps them
consolidate gear in their datacenters, it’s no surprise that startups are
lining up to drink from the multi-billion-dollar virtualization product
Count 3Leaf Systems as one of those in the line.
The company, which emerged from stealth mode in April with $32.5 million in
funding led by Intel Capital, launched
the V-8000 Virtual I/O Server.
This system, running 2 dual-core AMD Opteron processors, boosts the
efficiency of x86 servers by delivering I/O connectivity without excess
network and storage adapters, disks and switch ports in the data and storage
This machine is the first of what CEO and founder Bob Quinn said will be
several products that help business customers consolidate datacenter gear
and reduce capital expenditures and operational spending by 50 percent or
Quinn, who intends 3Leaf products to complement and support x86 computer
servers from IBM
, Sun Microsystems
, recently chatted with
internetnews.com about the company’s plans.
Q: What datacenter pain points is 3Leaf addressing?
Due to the transition from scale-up to scale-out computing over the last 10
years, we’ve had to deal with extremely low utilization on our servers. That
low utilization drives power and space, which really puts the datacenter in
a bind today.
Other elements that are painful are very long deployment
cycles. In this day and age, it should be possible, particularly with
virtualization, to deploy servers — maybe not instantly but certainly in under
an hour. Most datacenters take from four to 12 weeks to deploy a new server
with a new application stack. That is unacceptable.
Lastly, in the move away
from large mainframe and scale-up SMP systems to lower-cost but lower-efficiency x86-based servers, we really have comprised the reliability and
availability of our servers. That’s because the lower cost of x86 doesn’t
allow the same degree of reliability and availability with the far more
expensive scale-up systems.
Q: How does multi-core play into this?
We love multi-core because we view the core as the quantum of compute. This
plays directly into our long-term vision.
to enable your datacenter with commodity x86 servers, and then
we will deconstruct those servers logically and build a warehouse of
computer cores, memory and of I/O capability.
As we advance over the next
several years, we’ll go from 2 to 4 to 8 to 16 to 32 cores per commodity x86
microprocessor. This notion of deconstructing lets us view the datacenter in
a very different way.
Now when you want to launch a service or an
application stack on a particular operating system, provided it’s a
commodity operating system like a Windows or a Linux version, you can use
3Leaf’s products to power that operating system from beneath by resourcing
the right amount of cores, memory and I/O. This gives us an elastic
server, so you can write a policy telling that server to start with one
core. And when the business loads, indicate that one core will not be
sufficient, go grab another one from the warehouse to deliver the quality of
All of this is done while the system is running so there
is no interruption of service.
Q: How does this notion of the elastic server play into your long-term
Today, we look at virtualization as a way of slicing up a single server into
multiple virtual servers.
In this case, the problem we have in a production
environment is finding the right set of applications to put into that single
server such that we drive up the utilization but we don’t reach the point
where, for example, it’s multiples of that application peak at the same
time. You’ll run out of resources on that single server and you can no
longer deliver the quality of service.
In that case, the ideal server is one
huge enormous server to adjust the resources on demand. But large boxes are
really expensive. 3Leaf lets you build an enormous, but virtual box, which
is this pool of resources.
The first part of those resources is how we virtualize the I/O, which is the
product we just launched, the V-800. With the V-8000, we capture and
concentrate the I/O in an I/O server. The I/O has state but we effectively
take the I/O state away from the compute nodes, so the nodes are nothing but
processor and memory.
They’re then ripe for deconstructing and logically
taking their resources and putting them into a warehouse. In other words, the
core and memory have no meaning until it’s deployed as a resource for a
virtual operating system. We now dissociate the worldwide name from the
physical server and associate that with the worldwide name. This lets you
provision that state directly to a guest operating system, whether it’s an
OS running on bare metal or whether it’s running on a hypervisor, such as Xen
The second benefit is that in a typical server today, associated with gigabit
Ethernet or 2 or 4 gigabit Fiber Channel, those links are lightly used, so
we get the benefit of consolidation that lets us take 4 Fiber Channel ports
and use them to support 20 servers. So, we see a 5x to 10x consolidation in
terms of the I/O infrastructure required to support new servers.
There’s also an operational expenditure benefit, which is based on the
ability to give control of what server an application and operating system
runs on to the server administrator.
In today’s datacenter you can’t do that
because if you want to move from server A to server B, then you either have
your storage and network administrator cooperate to provision server B the
same way server A was provisioned, or you open up your views across all of
the boxes you want to move between. The enterprise isn’t prepared to do that
because there are tremendous security consequences associated with providing
views of all of your storage to your software engineers.
Q: What’s next for 3Leaf after the V-8000?
Our next follow-up is a beta launch of our compute memory virtualization
technology toward the end of this year. That will be game-changing, allowing
your datacenter to consist of warehouses of memory and cores to resources
running operating systems. This is your elastic server, where the resources
underneath the server will change based on the load presented to it.