Cloud computing data centers are production facilities that create and deliver networked services to customers, partners, employees, and regulators. Technology determines what they can do, but economics and the market determine what they must do—improve services despite ever-shrinking budgets, for example, or cope with explosive data growth without compromising service.

This new production environment fits a new set of demands.  New technologies make it possible to virtualize computing and storage, centralize infrastructure, and scale operations radically and instantly.  At the same time, businesses and consumers demand more, richer, and faster content and services delivered to any device, at any location.  Cloud computing matches these new capabilities and demands—-but only if data centers can overcome the challenges embedded in the design of today’s data-center networks.

Network complexity: what happened, and why it matters
With rare exceptions, today's data center networks have grown too costly and complex to realize the cloud computing vision.  Most IT executives will recognize these trends in their own data-center networks:

new application architectures—the old "stateful client-server" application model confined application traffic to “north/south” communications between servers and clients, but couldn't scale to serve tens of thousands of users, let alone millions.  Today's service-oriented and other Web 2.0 architectures spread that traffic across multiple specialized servers, placing new demands on the network.
new content—now that everything from training courses to phone calls is digitized, there's an exploding volume of progressively richer content to move and store.  And every innovation or performance improvement stimulates even more demand.

new clients—the explosion of portable and wireless clients, plus all the server and storage infrastructure needed to accommodate them, guarantees more network traffic and more complex network connections.

Attempts to stay ahead of these trends provide only short-term relief, and may even add to network complexity.  Examples include:
more network devices—the standard quick fix for capacity constraints, new network hardware just multiplies complexity as interconnections and workloads add latency and cost.

single purpose networks—data centers often address specific challenges by adding specialized networks.  Ethernet moves data, Fibre Channel SANs connect storage arrays to servers, and low-latency networks like InfiniBand carry inter-server communications. But network fragmentation greatly complicates both virtualization efforts and day-to-day network management.

concentration of resources and management—consolidating infrastructure and operations into owned, outsourced, or co-located data centers improves utilization, efficiency, and control, but raises the performance, availability, and security stakes for network infrastructure.

virtualization—essentially another way to concentrate resources, virtualization builds storage, servers, and network partitions out of fewer physical devices.  The approach lets multiple applications or user groups share physical infrastructure, for better utilization and elasticity from balancing workloads within resource pools.  But deployment and migration of virtual workloads place extreme demands on network performance and flexibility.

These problems aren’t unique to data center networks, but their symptoms show up first in the most demanding environments.  And they affect even companies with no plans to offer cloud computing services, but nevertheless rely on its component technologies.

The economics of simplifying your network
Today’s oversubscribed switching resources and sprawling aggregation and core layers are symptoms of legacy problems.  Most networks were originally designed to accommodate the limited throughput and port densities of an older generation of switches and routers, and the traffic volumes and patterns of client-server applications.

But networks no longer need to be designed around those deficiencies.  Ideally, a centralized, virtually partitioned network would be implemented as a single “logical” switch that is physically redundant (for availability), and capable of delivering high-performance, low-latency any-to-any conductivity across the data center for all traffic. 

Few of today’s data-center networks even approach this ideal, but most can profit from a fresh approach, using simplification as a core principle.  New technologies let designers reduce network complexity and total cost of ownership by collapsing switching layers, reducing interconnection links, consolidating security infrastructure, and standardizing on a single operating system to automate network operations across switching, routing and network-services infrastructure.  This simplified approach sets the stage for high performance today, and rapid adaptation and growth tomorrow, regardless of whether a company goes all the way to the highly-virtualized, network-intensive services of cloud computing.

The operating expense case is compelling.  Just reducing the number of switches cuts operating costs.  But once the network layer has been rationalized, economies crop up everywhere:
• Efficient design consolidates networking infrastructure into fewer devices, saving space, power, cooling, and management overhead.
• Flatter networks allow centralization and virtualization of security appliances, replacing multiple SSL, VPN, NAT and other separate devices.   Consolidating security services cuts expenses, and makes networks easier to manage, more efficient, and safer.
• Management efficiencies compound the savings.  Reducing the number and diversity of devices simplifies management.  Standards-based interfaces multiply the effectiveness of skilled staff, and support the use of automated tools for network management and routine technical support.

Performance in simplified networks
In addition to reducing deployment and management costs, flatter networks improve performance.  Every switch or device crossed by a data packet adds latency to the data path, so eliminating switching layers cuts end-to-end latency.  Flatter networks also help modern applications work better: Service Oriented Architecture (SOA) based applications, for example, benefit from data-center networks that can handle all the server-to-server traffic they generate.

High-performance, low-latency networks are particularly important for any company planning to scale its virtualization initiatives.  Network migration planning is an essential step toward virtualization at scale, and should be included in the very earliest stages.

Service and business improvements
Expense and performance improvements like these will be the immediate focus for organizations planning to unlock the benefits of new computing, storage, and software technologies.  But the benefits of a faster, simpler, more agile and less expensive foundation extend across the data center and the business it supports, for long-term operational improvements.

A single, highly scalable data center network makes it easier to pool server and storage resources, for more effective load balancing and better utilization.  It simplifies provisioning, reducing the time needed to introduce new services or execute a cycle of continuous improvement.  The bottom line is compelling: better quality services delivered in less time, at lower cost.

Whether your organization is planning a full-bore cloud computing initiative, or incremental gains from virtualization, networked storage, and new applications delivery models, take a good look at your network requirements.  Flatter, simpler network design will pay off quickly, and position you for quick adaptation to meet future challenges.

About the Author: Andy Ingram, vice president of product marketing and business development, Data Center Business Group, Juniper Networks. Andy has more than 25 years’ experience in the high-tech industry bringing ground-breaking technology to market.  In his current role, he is responsible for driving the marketing and go-to-market strategies for Juniper’s Data Center Business Group.  Andy joined Juniper in October 2008 from IGT, where he was the Senior VP of Network Systems.  Prior to IGT, Andy held various senior management positions at Sun Microsystems, Hewlett Packard, Cray Research and Sequent Computers, involved in the marketing and sales of servers, storage, system software, security products and application software.  Andy holds an MBA from the Anderson School at UCLA, a bachelor degree from the University of Colorado, and a CPA license in California.

Comments (0)

Subscribe to this comment's feed