Data Centers

Mapping Cisco Nexus, Catalyst, and MDS Logical Architectures

into Panduit Physical Layer Infrastructure Solutions
Introduction
The growing speed and footprint of data centers is challenging IT Managers to effectively budget and develop reliable,
high-performance, secure, and scalable network infrastructure environments. This growth is having a direct impact on
the amount of power and cooling required to support overall data center demands. Delivering reliable power and directly
cooling the sources that are consuming the majority of the power can be extremely difficult if the data center is not
planned correctly.

This design guide examines how physical infrastructure designs can support a variety of network layer topologies in
order to achieve a truly integrated physical layer infrastructure. By understanding the network architecture governing the
arrangement of switches and servers throughout the data center, network stakeholders can map out a secure and scalable
infrastructure design to support current applications and meet anticipated bandwidth requirements and transmission speeds.

The core of this guide presents a virtual walk through the data center network architecture, outlining the relationships of key
physical layer elements including switches, servers, power, thermal management, racks/cabinets, cabling media, and cable
management. The deployment of Cisco hardware in two different access models are addressed: Top of Rack (ToR) and
End of Row (EoR).

Top of Mind Data Center Issues Developing the Integrated

The following issues are critical to the process of building Infrastructure Solution
and maintaining cost-effective network infrastructure Data center planning requires the close collaboration of
solutions for data centers: business, IT, and facilities management teams to develop
an integrated solution. Understanding some general planning
• Uptime relationships helps you translate business requirements into
Uptime is the key metric by which network reliability practical data center networking solutions.
is measured, and can be defined as the amount of
time the network is available without interruption
to application environments. The most common
service interruptions to the physical layer result
from operational changes.

• Scalability
When designing a data center, network designers
must balance today’s known scalability requirements
with tomorrow’s anticipated user demands. Traffic
loads and bandwidth/distance requirements will
continue to vary throughout the data center,
which translates to a need to maximize your
network investment.

• Security
A key purpose of the data center is to house
mission critical applications in a reliable, secure
environment. Environmental security comes in
many forms, from blocking unauthorized access to
monitoring system connectivity at the physical layer.
Overall, the more secure your network is, the more
reliable it is.

2
Business Requirements Drive Data Center Design
A sound data center planning process is inclusive of the needs of various business units. Indeed, the process requires the
close collaboration of business, IT, and facilities management teams to develop a broad yet integrated data center solution
set. Understanding some general planning relationships helps you translate business requirements into practical data center
networking solutions.

Business requirements ultimately drive all data Business Stakeholders

center planning decisions (see Figure 1). On

a practical level, these requirements directly
impact the type of applications deployed and
Business Requirements
Service Level Agreements (SLAs) adopted by the
organization. Once critical uptime requirements
are in place and resources (servers, storage,
compute, and switches) have been specified to
support mission-critical applications, the required Service Level Applications
Agreement Software
bandwidth, power, and cooling loads can
be estimated.
Data Center
Solution

Some data center managers try to limit the

number of standard compute resources on fewer
hardware platforms and operating systems, which Active Equipment
Power Requirements
(Servers, Storage)
makes planning decisions related to cabinet, row, Structured Cabling
and room more predictable over time. Other
managers base their design decisions solely on Network Requirements
Cooling Requirements
the business application, which presents more (Switching)

of a challenge in terms of planning for future

growth. The data center network architecture Facilities Stakeholders IT Stakeholders

discussed in this guide uses a modular, End of Row

(EoR) model and includes all compute resources Figure 1. Business Requirements
and their network, power, and storage needs. The
resulting requirements translates to multiple LAN, SAN, and power connections at the physical layer infrastructure.
These requirements can be scaled up from cabinet to row, from row to zone, and from zones to room.

Designing Scalability into the Physical Layer

When deploying large volumes of servers inside the data One way to simplify the design and simultaneously
center it is extremely important that the design footprint is incorporate a scalable layout is to divide the raised floor
scalable. However, access models vary between each space into modular, easily duplicated sub-areas. Figure
network, and can often be extremely complex to design. 2 illustrates the modular building blocks used in order to
The integrated network topologies discussed in this guide design scalability into the network architecture at both OSI
take a modular, platform-based approach in order to scale Layers 1 and 2. The logical architecture is divided into three
up or down as required within a cabinet or room. It is discrete layers, and the physical infrastructure is designed
assumed that all compute resources incorporate resilient and divided into manageable sub-areas called “Pods”. This
network, power, and storage resources. This assumption example shows a typical data center with two zones and 20
translates to multiple LAN, SAN, and power connections Pods distributed throughout the room; core and aggregation
within the physical layer infrastructure. layer switches are located in each zone for redundancy, and
access layer switches are located in each Pod to support the
compute resources within the Pod.

3
 C O LD A IS LE DC

Pod Zone
H O T A IS LE
Pod

N e tw o rk R a ck

S e rve r R a ck

S to ra g e R a ck

Pod

Module 1 Module N

Po d Po d

Figure 2. Mapping the Logical Architecture to the Cabling Infrastructure

Network Access Layer Environments

This guide describes two models for access layer switching environments – Top of Rack (TOR), and End of Row (EOR) – and
reviews the design techniques needed for the successful deployment of these configurations within an integrated physical
layer solution. When determining whether to deploy a TOR or EOR model, it is important to understand the benefits and
challenges associated with each:

• A ToR design reduces cabling congestion which enhances flexibility of network deployment and installation. Some
trade-offs include reduced manageability and network scalability for high-density deployments due to the need to
manage more access switches than in an EoR configuration

• An EoR model (also sometimes known as Middle of Row [MoR]) leverages chassis-based technology for one or
more row of servers to enable higher densities and greater scalability throughout the data center. Large modular
chassis such as the Cisco Nexus 7000 Series and Cisco Catalyst 6500 Series allow for greater densities and
performance with higher reliability and redundancy. Figure 2 represents an EoR deployment with multiple Pods
being distributed throughout the room

Note: Integrated switching configurations, in which applications reside on blade servers that have integrated switches built into each chassis, are not
covered in this guide. These designs are used only in conjunction with blade server technologies and would be deployed in a similar fashion as EoR
configurations.

4
Top of Rack (ToR) Model
The design characteristic of a ToR model is the inclusion of an access layer switch in each server cabinet, so the physical
layer solution must be designed to support the switching hardware and access-layer connections. One cabling
benefit of deploying access layer switches in each server cabinet is the ability to link to the aggregation layer using
long-reach small form factor fiber connectivity. The use of fiber eliminates any reach or pathway challenges presented
by copper connectivity to allow greater flexibility in selecting the physical location of network equipment.

Figure 3 shows a typical logical ToR network topology, Nexus

illustrating the various redundant links and distribution of 7010
connectivity between access and aggregation switches. This
example utilizes the Cisco Nexus 7010 for the aggregation layer
and a Cisco Catalyst 4948 for the access layer. The Cisco
Catalyst 4948 provides 10 GbE links routed out of the cabinet
back to the aggregation layer and 1 GbE links for server access
connections within the cabinet.

Once the logical topology has been defined, the next step is Catalyst
to map a physical layer solution directly to that topology. 4948
With a ToR model it is important to understand the number
of network connections needed for each server resource. The
basic rule governing the number of ToR connections is that any
server deployment requiring more than 48 links requires an
additional access layer switch in each cabinet to support the
higher link volume. For example, if thirty (30) 1 RU servers
that each require three copper and two fiber connections
are deployed within a 45 RU cabinet, an additional access
layer switch is needed for each cabinet. Figure 4 shows the
typical rear view ToR design including cabinet connectivity Figure 3. Logical ToR Network Topology
requirements at aggregation and access layers.

Patch panel Patch panel

Top of Rack Top of Rack
Patch panel Patch panel
server server
x – connect x – connect

Network Network
Aggregation Aggregation
Point Point
A–B A–B
server server

Figure 4. Rear View of ToR Configuration

5
Density considerations are tied to the level of
redundancy deployed to support mission critical
hardware throughout the data center. It is
critical to choose a deployment strategy
that accommodates every connection
and facilitates good cable management.

High-density ToR deployments like the one shown SAN

Connections
in Figure 5 require more than 48 connections
per cabinet. Two access switches are deployed
in each Net-Access™ N-Type Network Cabinet to
LAN
support complete redundancy throughout the Connections
network. All access connections are routed within
the cabinet and all aggregation linked are routed up
and out of the cabinet through the FiberRunner ®
Cable Routing System back to the horizontal Power
distribution area. Connections

Lower-density ToR deployments require less

than 48 connections per cabinet (see Figure 6).
This design shares network connections between Figure 5. Dual Switch – Server Cabinet Rear View
neighbor cabinets to provide complete redundancy
to each compute resource.

Figure 6. Single Switch – Server Cabinet Rear View

6
The cross-over of network connections between cabinets the lower portion of the cabinet closer to the cooling source
presents a cabling challenge that needs to be addressed in to allow for proper thermal management of each compute
the physical design to avoid problems when performing any resource. In this layout, 1 RU servers are specified at
type of operational change after initial installation. To properly 24 servers per cabinet, with two LAN and two SAN
route these connections between cabinets there must connections per server to leverage 100% of the LAN
be dedicated pathways defined between each cabinet to switch ports allocated to each cabinet.
accommodate the cross-over of connections. The most
common approach is to use Panduit overhead cable Connectivity is routed overhead between cabinets to
routing systems that attach directly to the top of the minimize congestion and allow for greater redundancy
Net-Access™ N-Type Network Cabinet to provide dedicated within the LAN and SAN environment. All fiber links from
pathways for all connectivity routing between cabinets, as the LAN and SAN equipment are routed via overhead
shown in Figure 6. pathways back to the Cisco Nexus, Catalyst, and MDS
series switches at the aggregation layer. The Panduit ®
Table 1 itemizes the hardware needed to support a typical FiberRunner ® Cable Routing System supports overhead
ToR deployment in a data center with 16,800 square feet fiber cabling, and Panduit ® Net-Access™ Integral Cabinet
of raised floor space. Typically in a ToR layout, the Cisco Top Cable Routing System can be leveraged with horizontal
Catalyst 4948 is located towards the top of the cabinet. cable managers to support copper routing and patching
This allows for heavier equipment such as servers to occupy between cabinets.

Data Center Assumptions Quantity Server Specifications:

Raised Floor Square Footage 16800 2 — 2.66 GHz Intel Quad Core Xeon X5355
Servers 9792 2 x 670 W Hot-Swap
Cisco Nexus 7010 10 4 GB Ram — (2) 2048MB Dimm(s)
Cisco Catalyst 4948 408 2 — 73GB 15K-rpm Hot-Swap SAS — 3.5
Server Cabinets 408
Network Cabinets (LAN and SAN Equipment) 33
Midrange and SAN Equipment Cabinets 124

Watts Per Device Per Cabinet Cabs Per Pod Pods Per Room Quantity Total Watts
Servers 350 24 102 4 9792 3,427,200
Cisco Catalyst 4948 Switches (Access) 350 1 102 4 408 142,800
Cisco MDS (Access) 100 1 102 4 408 40,800
Cisco Appliance Allocation 5,400 — — — 8 43,200
Cisco Nexus 7010 (Aggregation) 5,400 — 2 4 8 43,200
Cisco MDS (Aggregation) 5,400 — 2 4 8 43,200
Cisco Nexus 7010 (Core) 5,400 — — — 2 10,800
Cisco MDS (Core) 5,400 — — — 2 10,800
Midrange/SAN Equipment Cabinets 4,050 — — — 124 502,200
Midrange/SAN Switching MDS 5,400 — — — 4 21,600
Midrange/SAN Switching Nexus 7010 5,400 — — — 4 21,600
Total Watts: 4,307,400
Table 1. Room Requirements for Typical ToR Deployment Total Kilowatts: 4,307.40
Total Megawatts: 4.31

End of Row (EoR) Model

In an EoR model, server cabinets contain patch fields but not access switches. In this model, the total number of servers
per cabinet and I/Os per server determines the number of switches used in each Pod, which then drives the physical layer
design decisions. The typical EoR Pod contains two Cisco Nexus or Cisco Catalyst switches for redundancy. The length of
each row within the Pod is determined by the density of the network switching equipment as well as the distance from the
server to the switch. For example, if each server cabinet in the row utilizes 48 connections and the switch has a capacity
for 336 connections, the row would have the capacity to support up to seven server cabinets with complete network
redundancy, as long as the seven cabinets are within the maximum cable length to the switching equipment.

7
 Figure 7. Top View of EoR Configuration

Figure 7 depicts a top view of a typical Pod design for a EoR configuration, and shows proper routing of connectivity from
a server cabinet to both access switches. Network equipment is located in the middle of the row to distribute redundant
connections across two rows of cabinets to support a total of 14 server cabinets. The red line represents copper LAN “A”
connections and the blue line represents copper LAN “B” connections. For true redundancy the connectivity takes two
diverse pathways using Panduit ® GridRunner™ Underfloor Cable Routing Systems to route cables underfloor to each cabinet.
For fiber connections there is a similar pathway overhead to distribute all SAN “A” and “B” connections to each cabinet.

As applications continue to put even greater demands on

the network infrastructure it is critical to have the appropriate
cabling infrastructure in place to support these increased
bandwidth and performance requirements. Each EoR-arranged
switch cabinet is optimized to handle the high density
requirements from the Cisco Nexus 7010 switch. Figure 8
depicts the front view of a Cisco Nexus 7010 switch in a
Net-Access™ N-Type Network Cabinet populated with Category
6A 10 Gigabit cabling leveraged for deployment in an EoR
configuration. It is critical to properly manage all connectivity
exiting the front of each switch. The EoR-arranged server cabinet
is similar to a typical ToR-arranged cabinet (see Figure 5), but
the characteristic access layer switch for both LAN and SAN
connections are replaced with structured cabling.

Table 2 itemizes the hardware needed to support a typical EoR

deployment in a data center with 16,800 square feet of raised
floor space. The room topology for the EoR deployment is not
drastically different from the ToR model. The row size is
determined by the typical connectivity requirements for any
given row of server cabinets. Most server cabinets contain
a minimum of 24 connections and sometimes exceed 48
connections per cabinet. All EoR reference architectures
are based around 48 copper cables and 24 fiber strands
for each server cabinet. Figure 8. Panduit® Net-Access™ N-Type Network
Cabinet with Cisco Nexus Switch

8
Data Center Assumptions Quantity Server Specifications:
Raised Floor Square Footage 16800 2 — 2.66 GHz Intel Quad Core Xeon X5355
Servers 9792 2 x 835 W Hot-Swap
Cisco Nexus 7010 50 16 GB Ram — (4)4096MB Dimm(s)
Server Cabinets 336 2 — 146GB 15K-rpm Hot-Swap SAS — 3.5
Network Cabinets (LAN and SAN Equipment) 108
Midrange and SAN Equipment Cabinets 132
Watts Per Device Per Cabinet Cabs Per Pod Pods Per Room Quantity Total Watts
Servers 450 12 14 24 9792 1,814,400
Cisco Appliance Allocation 5,400 — — — 8 43,200
Cisco Nexus 7010 (Aggregation) 5,400 — 2 24 48 259,200
Cisco MDS (Access) 5,400 — 2 24 48 259,200
Cisco Nexus 7010 (Core) 5,400 — — — 2 10,800
Cisco MDS (Core) 5,400 — — — 2 10,800
Midrange/SAN Equipment Cabinets 4,050 — — — 132 502,200
Midrange/SAN Switching MDS 5,400 — — — 4 21,600
Midrange/SAN Switching Nexus 7010 5,400 — — — 4 21,600
Total Watts: 4,307,400
Table 2. Room Requirements for Typical EoR Deployment Total Kilowatts: 4,307.40
Total Megawatts: 4.31

End of Row (EoR) Model

Panduit is focused on providing high-density, flexible physical layer solutions that maximize data center space utilization
and optimize energy use. The following sections describe cabinet, cooling, and pathway considerations that are common to
all logical architectures.

Cabinets
Cabinets must be specified, that allow for maximum scalability and flexibility within the data center. The vertically mounted
patch panel within the cabinet provides additional rack units that can be used to install more servers within the 45 rack units
available. These vertically mounted panels also provide superior cable management versus traditional ToR horizontal patch
panels by moving each network connection closer to the server network interface card ultimately allowing a shorter patching
distance with consistent lengths throughout the cabinet.

Figure 9 represents a typical vertical patch panel deployment in

the Net-Access™ N-Type Network Cabinet. Complete power and
data separation is achieved through the use of vertical
cable management on both sides of the cabinet frame.
The cabinet also allows for both overhead and underfloor
cable routing for different data center applications. Shorter
power cords are an option to remove the amount of added
cable slack from longer cords that are shipped with server
hardware. Using shorter patch cords and leveraging Panduit
vertical cable management integrated into the Net-Access™
N-Type Network Cabinet alleviates potential airflow issues
at the back of each server that could result from poorly
managed cabling.

Figure 9. Cabinet Vertical Mount Patch Panel

9
Thermal Management management systems that efficiently route and organize
Data center power requirements continue to increase at high critical IT infrastructure elements. Figure 10 represents an
rates making it difficult to plan appropriately for the proper analysis done based upon the assumptions for the 16,800
cooling systems needed to support your room. Cabinets play square foot EoR deployment.
a critical role in managing the high heat loads generated by
active equipment. Each cabinet will require different power Panduit ® Net-Access™ N-Type Network Cabinets feature
loads based upon the type of servers being installed as well large pathways for efficient cable routing and improved
as the workload being requested of each compute resource. airflow while providing open-rack accessibility to manage,
Understanding cabinet level power requirements gives protect, and showcase cabling and equipment. Elements
greater visibility into overall room conditions. such as exhaust ducting, filler panels, and the Panduit®
Cool Boot® Raised Floor Assembly support hot and cold
Panduit Laboratories’ research into thermal management aisle separation in accordance with the TIA-942 standard.
includes advanced computational fluid dynamics (CFD) These passive solutions (no additional fans or compressors)
analysis to model optimal airflow patterns and above-floor contribute by minimizing bypass air in order to manage
temperature distributions throughout the data center. This higher heat loads in the data center and ensure proper
data is then used to develop rack, cabinet, and cable equipment operation and uptime.

Designed in a Hot – Cold Architecture

12 – 20 Ton CRAC Units Outside

12 – 30 Ton CRAC Units Inside

Utilizing Ceiling Plenum for

Return Air

All Perforated Tiles at 25% Open

Peak Temp was 114°

Figure 10. Data Center EoR Thermal Analysis

Pathways This strategy offers several benefits:

The variety and density of data center cables means that
there is no “one size fits all” solution when planning cable • The combination of overhead fiber routing system
pathways. Designers usually specify a combination of and cabinet routing system ensures physical
pathway options. Many types and sizes are available for separation between the copper and fiber cables, as
designers to choose from, including wire basket, ladder rack, recommended in the TIA-942 data center standard
J-hooks, conduit, solid metal tray, and fiber-optic cable
routing systems. Factors such as room height, equipment • Overhead pathways such as the Panduit®
cable entry holes, rack and cabinet density, and cable types, FiberRunner ® Cable Routing System protect
counts, and diameters also influence pathway decisions. fiber optic jumpers, ribbon interconnect cords, and
The pathway strategies developed for ToR and EoR models multi-fiber cables in a solid, enclosed channel that
all leverage the FiberRunner® Cable Routing System to route provides bend radius control, and the location of the
horizontal fiber cables, and use the Net-Access™ Integral pathway is not disruptive to raised floor cooling
Cabinet Top Overhead Cable Routing System in conjunction
with Wyr-Grid® Wire Basket or ladder rack for horizontal • The overall visual effect is organized, sturdy,
copper and backbone fiber cables. and impressive

10
Conclusion
Next-generation data center hardware such as the Cisco
Nexus 7010 switch provides increased network capacity
and functionality in the data center, which in turn is
placing greater demands on the cabling infrastructure.
This guide describes the ways that Panduit structured
cabling solutions map easily to the logical architectures
being deployed in today’s high-performance networks
to achieve a unified physical layer infrastructure.
Network stakeholders can use modular designs for
both hardware architectures and cabling layouts to
ensure that the system will scale over the life of the
data center to survive multiple equipment refreshes
and meet aggressive uptime goals.

About Panduit
Panduit is a leading, world-class developer and provider
of innovative networking and electrical solutions.
For more than 50 years, Panduit has engineered and
manufactured end-to-end solutions that assist our
customers in the deployment of the latest technologies.
Our global expertise and strong industry relationships
make Panduit a valuable and trusted partner dedicated
to delivering technology-driven solutions and unmatched
service. Through our commitment to innovation, quality
and service, Panduit creates competitive advantages to
earn customer preference.

Cisco, Cisco Systems, Nexus, Catalyst, and MDS are

registered trademarks of Cisco Technology, Inc.

www.panduit.com • [email protected] • 800-777-3300

©2013 Panduit Corp.

ALL RIGHTS RESERVED.
CPCB116--WW-ENG
10/2013

11