Management, and Control of IT Infrastructures

Networks are wonderful things. They connect servers, resources, applications, and
users together, all to drive your business forward. But as application complexity goes
up, as the need for more and more servers goes up, as more and more virtual servers
are deployed, and as the need for more and more bandwidth increases, networking
complexity increases as well.
There are routers and switches, NICs and more NICs, and cables and cables, and more
cables. Oh, the cables! Hundreds, even thousands of wires are running all over your
data center. And if you want to change a server or reconfigure an application, the time
and effort involved in juggling cables is almost ridiculous.
Cables and networks should help your business flow. Instead, as your business grows,
they become complex and challenging bottlenecks to productivity.
Fortunately, all that is about to change.
 
The Network Nightmare
The modern data center is a networked data center. Virtual servers live in physical
servers and blades. Blades live in racks. And all those virtual servers, all those physical
servers and blades, and all those racks need to talk to each other:

NICs, Switches, and Cables
When data leaves a server, it does so through a network interface card 
(NIC), then travels across a cable to a switch, which acts like a 
switchboard, then travels to and across other switches, and eventually 
travels through another NIC back into another server. Each server uses 
one or more NICs. As more servers are added, more NICs are added. As 
more NICs are added, more switches are needed. Connecting them all are 
cables. 
Virtualization Magnifies the Problem 
Virtualization reduced the number of physical boxes required for certain 
applications, but it didn’t reduce the network load. In fact, virtual servers 
have substantially increased the network load. It’s now almost absurdly 
easy to add a new server, but terribly difficult to add the network and cable 
infrastructure to support the new server. 
LANs vs. SANs 
Data center data and storage traffic  both travel across cables. Ethernet 
(and most IP-based applications) can  survive having some packets lost 
and then re-sent during  operation. But storage networking must be 
lossless, and must maintain throughput.

These separate levels of required performance have given rise to two completely divergent 

network infrastructures, one based mostly on Ethernet, and the other based mostly 

on Fiber Channel.  

The result is once again more cables, more connections, less flexibility, 

and increased complexity. 

Cable Aggregation  

Racks often contain many more servers today than they did even a few 

years ago. Virtualization has made  that possible. Unfortunately, the 

virtualization trend didn’t reduce the need for more wires. Instead it made 

it worse. 

Wiring and provisioning racks require many complex connections be made 

at top-of-rack, and then all of those connections are often bundled 

together at another set of switches at end-of-row. Always, the complexity 

increases.

It’s reached the point where the task of adding a completely new server application can
take far less time than simply wiring up the connections.  Gartner's Cameron Haight
reports that setting up a virtual machine is relatively fast (a matter of hours), but
provisioning all the networking changes required to support that  virtual machine can
take as long as six weeks.
Networking, a technology designed to help businesses grow, is instead getting in the
way of growth. IT organizations are change-adverse because a simple change, like
adding one new application, could have a complex ripple effect throughout the entire
data center.
The number of connections, switches, and cables keeps going up.  IT staff is spending
more and more time on maintenance and wires. The technical wizards who should be
driving innovation across the organization are, instead, being driven to distraction simply
attempting to keep up.
The Network as a Service 
The solution is turning the network into  a service, creating a virtualized network
infrastructure. What makes all this work  is a network topology known as a switched
fabric. Rather than using a lot of point-to-point connections, a fabric-based network
allows many nodes to connect to each other like threads in a tapestry, often dynamically
changing to compensate for changing load requirements.
Once you’re able to dynamically reallocate your network resources, you’re able to begin
serving network resources on-demand, both when and where they’re needed.

Here’s how:
Virtualize Connections 
Typically, one NIC supports one connection. Each new connection 
requires more NICs. It’s now possible, however, to virtualize your network 
connections, so each NIC can support multiple connections. You can 
dynamically add more connections and change the purpose of those 
connections on-the-fly.
Wire Once 
Once you virtualize your connections, you can build out your physical 
wiring infrastructure once. When applications and server requirements 
change, you can make changes in your virtualized connections through 
management software, eliminating the need to pull wires for each change 
in application or server requirements.
Repurpose Without Tears 
Once you embrace the concept of wiring once, you can repurpose 
network, storage, and compute resources on the fly. Your IT department is 
no longer subject to waiting days or weeks for a re-provisioned network. 
Instead, the network can be modified  to meet line-of-business needs 
quickly, and without drastic and complex changes to the physical data 
center facilities.
Increase Performance 
Complexity breeds bottlenecks. The more connections, wires, and 
switches you have, the more overhead you have running across your 
network, and the slower everything gets. 
Once you can virtualize your network infrastructure, you can cut 90% or 
more of your physical wiring, which will substantially drop the network 
housekeeping overhead. The result is increased performance, often by as 
much as 50%.

It’s astonishing how much IT time and effort is bottled up in cabling bottlenecks. By
virtualizing the network and turning your network into a service, most of that time can be
freed up and put to far better use.

Organizational Benefits 
When you transition to the network as a service, you’ll be taking complexity out of your
network. This will make it easier for you to manage not just basic networking elements,
but your entire IT infrastructure.
You’ll be able to turn your IT operation into an agile, responsive, innovative contributor
to your organization’s mission. You’ll also able to do more, even while spending less on
expensive networking gear. You can allocate what spending you do to support line-ofbusiness
objectives rather than for basic, repetitive, and inefficient operations.
You’ll be able to maximize your data center space, reduce your overall power costs, and
even free up strategic IT personnel for more bottom-line oriented tasks.
Future Innovations 
Today, network fabrics still have one substantial limitation: LAN vs. SAN. Local Area
Networks are usually Ethernet-based, while Storage Area Networks run on Fiber
Channel. Even though virtualized networks remove most of the cabling overhead,
there’s still the need for these two types of cables.
That’s changing. Standards bodies are working on encapsulating Fiber Channel frames
into Ethernet frames, so that all of the SAN traffic can run across Ethernet. This
technology is called Fiber Channel over Ethernet, or FCoE.
But encapsulating frames won’t solve the bigger issue, the need for a lossless Ethernet
mechanism, especially across a congested data center network. The Data Center
Bridging (DCB) Task Group of the IEEE 802.1 Working Group is working on adding
performance capabilities to Ethernet called Converged, Enhanced Ethernet.
Some venders are currently offering solutions they claim will be compliant with these
future standards. If you decide to implement these solutions now, make sure your
performance agreements include upgrades for  pre-standard revisions and eventually,
final standards compliance.
In the very near future, you’ll be able to  transition your network to a policy-based
approach for managing your infrastructure. You’ll be able to control it as a resource,
reallocating workloads dynamically depending on your needs at the time, whether
across the data center or even between data centers.