By Stephen J. Bigelow, Senior Technology Writer
As you saw in the first part of this Hot Spot Tutorial, solution providers can deploy basic networks using inexpensive, readily available and highly commoditized network elements like switches, NICs, network storage systems, routers and so on. But simply deploying a series of basic components is not always enough. Networks are getting bigger and the demands for network access, performance and uptime are increasing. Solution providers must implement network designs that meet the client's needs for performance and reliability -- minimizing downtime and maintaining network flexibility. This second installment of our network design tutorial discusses some of the most important considerations for robust network design.
Robust network design considerations
One of the first considerations in robust network design is resilience (or resiliency). Resilience is the ability to sustain an acceptable level of service in the event of faults, changes or other challenges to normal network operation. Thus, a resilient network can recover from hardware, software or process problems.
Network resilience can take several forms depending on which solution provider you talk to. In some cases, the challenge is change, and some solution providers see resilience as a matter of network flexibility -- network architecture choices that can accommodate future technology changes or additional load. "What level of demand can you serve with the infrastructure that you have?" said Dave Sobel, CEO of Evolve Technologies, a solution provider headquartered in Fairfax, Va.
Resilience is often intertwined with network redundancy. A redundant network design has secondary (or even tertiary) resources that can maintain functionality when the primary element fails. For example, a client that depends on Internet access may require redundant WAN connectivity. This may employ WAN providers that use entirely different physical circuits, ensuring that access is available even when one of the connections (or ISPs) is disrupted. Similarly, network infrastructures may include redundant links between major network segments, so portions of the network will remain available even when one link goes down.
The main issues with redundancy in network design are risk and cost justification for the client. Adding components raises the cost of a network design project. "You only really see redundancy in larger networks," said Adam Gray, chief technology officer of Novacoast, an IT professional services company in Santa Barbara, Calif. "Many smaller companies have the ability to forego redundancy if the cost outweighs the risk. In larger environments, you just don't have that option."
Trunking and failover also play a role in robust network design. Trunking is the aggregation of multiple links which can be used together to increase the effective data throughput between trunked points within the network. Failover is the orderly transfer (or handoff) of operation from one network device or link to another, allowing traffic to continue even in the face of a failure. Trunking is a key element of failover, allowing redundant communication paths between network devices. While failover is important for most organizations, trunking has not been critical for added throughput due to the availability of fast network technologies like 1 Gigabit Ethernet (GbE).
As an integral part of the client's infrastructure, a storage area network (SAN) can benefit from director-class switches. Director switches provide reliable high-speed connections and support nonblocking port capability so that all ports operate at full speed. A solution provider may consider adding or upgrading to a director switch when the client's current SAN switch runs low on ports or needs better performance, or the client needs to tie together more storage systems or SAN "islands." Director switches also feature storage traffic management and storage virtualization capabilities. Similarly, iSCSI storage networks can be improved by adding high-performance, low-latency switches that are optimized for storage tasks.
TCP/IP packet processing has always been handled by the CPU. The move to 1 GbE and higher, however, can place a strain on processing resources -- especially virtualized processing resources (virtual servers) that are supporting multiple simultaneous workloads, or time- and latency-sensitive data like iSCSI traffic. TCP/IP Offload Engine (TOE) cards reduce CPU load on heavy network communication loads.
"Different manufacturers provide different levels of functionality, from simple checksum offload to more complex packet processing," said Allen Zuk, an independent IT risk management consultant formerly with GlassHouse Technologies. "Some couple the hardware offload function with special driver software to further improve the flow of data." The principal disadvantage of TOE cards is their expense. At $500 or more per port, solution providers will have to make a compelling financial argument to the client.
Finally, any network design project should include an evaluation of the client's cabling infrastructure. Most Ethernet cabling is classified as Category 5 -- suitable for network traffic up to 1 GbE. Faster 10 GbE network deployments will require more recent Category 6 cabling, so solution providers will need to consider the client's ultimate goals. For example, solution providers installing new cable runs should consider using Category 6 by default, allowing faster networking deployments in the future without the hassle of massive recabling costs. "I think if you're a small-business reseller, you're dealing with cabling a lot more," Gray said, noting that larger clients can usually get cabling jobs done by firms that specialize in that work.
Larger clients that rely on WAN connectivity to branch offices or remote locations may add WAN acceleration technologies to the network. WAN acceleration improves bandwidth efficiency and reduces remote application response times or latency. While solution providers recognize the benefits of acceleration, continued advances in network technology are applying pressure against WAN acceleration.
"Bandwidth has gotten much cheaper," Gray said, noting that cheaper bandwidth eases the need for WAN acceleration, and pointing to the emergence of other technology offerings from telecommunication providers, such as Multi Protocol Label Switching (MPLS). "We're seeing far fewer clients invest in WAN acceleration and most of them investing in the MPLS cloud architecture," Gray added. Solution providers should consider adding WAN acceleration to the network design project if the cost of additional bandwidth (or lost productivity) is greater than the expense of the acceleration product.
It's important to note that robust network design choices are not necessarily top priorities for the client. This is particularly true with the client's upper management, which is mainly concerned with business issues such as cost, risk and return on investment. Focus client discussions on business goals and needs assessment, and then explain how robust network design elements will help the client achieve those higher goals. For example, no solution provider sits down with a client to discuss failover, but the solution provider may present a project plan that recommends using failover with redundant connections -- which the provider recognizes as mission-critical for the client.
Measuring and adding performance in network designs
Network design projects use bandwidth and performance monitoring during the pre-assessment phase to examine performance levels, identify bandwidth utilization patterns and establish a performance baseline. Monitoring is then used after the project deployment to gauge results and verify successful completion of design goals. There are numerous network monitoring tools available, each focusing on various hardware or software attributes of the network. You can find a comprehensive list of network monitoring tools at the Stanford Linear Accelerator Center (SLAC) website.
A network design project may involve adding bandwidth on the WAN (Internet) or LAN side of the router. The driving factors are subjective -- data transfers may take too long, application performance may appear erratic and so on. Monitoring tools are important because they quantify the performance of network elements in tangible and repeatable terms. Most testing is performed on subsections of the network first, then narrowed to focus on potential problem areas. "Typically segments are targeted first as it is cheaper to add faster switch ports than to upgrade a network circuit," Zuk said.
Solution providers versed in troubleshooting may recognize monitoring patterns that lead to configuration improvements or device upgrades that have little to do with bandwidth. Still, heavy utilization patterns are the best justification for adding bandwidth. "The time to add more bandwidth is when your line is saturated on a consistent basis and it's impacting performance," Sobel said.