A popular stance in network design is wireless connectivity should augment but not replace wired connections, primarily because of some of the disadvantages of wireless networking. Yet, wireless LAN applications continue to mature in features and usability. Higher speeds, increased security, quality of service (QoS) and centralized management are just a few of the wireless developments in the past few years, and more advances are coming. Does this mean that the time has come to completely abandon the traditional design of a wired port to every desktop? Certainly, there are numerous successful wireless only deployments out there.
Determining whether a network should be wired, wireless or a mixture of both should be part of every new network design process. Often a decision on how to proceed is based on what worked well the year before for a similar project. However, since the available offerings change rapidly, the question of wired vs. wireless should be explored in-depth as part of every network design process, regardless of what worked for a previous project of similar scope.
I recently had the task of designing a LAN for a new addition to an existing medical building, and I went through the process of determining what application worked best for that environment. While doing so, I carefully considered technologies and trends, network needs and security, and cost and management factors.
Technologies and trends
Understanding the available and future technologies is necessary in designing a network. In widespread deployment today are 100Mbit/sec. and, to a lesser extent, 1Gbit/sec. Ethernet over twisted-pair copper cabling to the desktop. On the wireless side, 802.11b has proven to be the workhorse of WLAN connectivity, with 802.11a and 802.11g providing higher speeds for that freedom.
The need for bandwidth is ever-growing, particularly in medical, financial, advanced computing and research environments. To address these needs, the IEEE has focused on expanding both wireless and wired capabilities. 802.3an was approved last year as a standard for providing 10Gbit/sec. Ethernet over copper cabling, and recently the IEEE 802.3 Higher Speed Study Group announced it will focus on developing a 100Gbit/sec. Ethernet over copper standard.
On the wireless side, according to the 802.11 Official Timelines, 802.11n promises up to 540Mbit/sec. throughput and is projected for approval as a standard in April 2008. Devices based on the 802.11n draft, such as the Linksys Wireless-N Broadband Router, are currently on the market. However, installing based on a standard before it is ratified can potentially introduce future interoperability issues.
The important concept to keep in mind is that while the network equipment industry is producing products for both media with increased bandwidth capacity, wired will most likely continue to hold a significant edge in throughput. The question then that needs to be addressed is whether the added bandwidth is needed by the projected uses of the network and is addressed in the next section.
Convergence is a trend that will continue. Voice over IP (VOIP) is no longer a wired-only option, with softphones and handheld 802.11b sets, such as the Cisco Unified Wireless IP Phone 7920, currently available. Many access points offer QoS to ensure the small bandwidth needed per call is available. Voice terminals — wired or wireless — don’t need much bandwidth per call (64Kbit/sec.) but demand low latency. A QoS configured access point can ensure someone downloading a large file will not interfere with a sales call to a client.
Convergence does not end with VOIP. Roaming between IP and cellular networks is already offered in a pilot project by T-Mobile in limited areas. So long as the service is sound and the technology matures, it is reasonable to expect offerings of this type of converged services to expand.
Keeping up with the developments in technology and trends is important, but only if applied properly. To do so, a clear understanding of the needs of the network users is necessary.
Network needs and security
When designing a network, it’s necessary to consider what type of data the network will transmit. A network is simply a tool, and an old mechanic’s axiom is you must use the right tool for the job. A thorough analysis of such uses will ensure the performance of the network is satisfactory.
Bandwidth needs are dictated by the applications that will use the network. Two application specific bandwidth requirements need to be considered: throughput and latency. Throughput is the data transmission rate, measured in bits per second, whereas latency is the delay, or lag, in data transmission.
Some applications such as software downloads, Web browsing and e-mail work well with some amount of latency because where the induced lag is not noticeable. As noted above, “real time” type applications, such as VOIP, don’t require a large amount of throughput but do demand low latency. Even where data transfer is heavy, such as with online backups where high throughput is necessary, some latency is acceptable.
One user connected to an access point at 54Mbit/sec. that has a hardwired 100Mbit/sec. switched connection will most likely experience acceptable performance. But all wireless designs are not equal, and several factors will decrease the end user’s perceived performance.
Access point connectivity is a shared network model. One user connected to an access point downloading a Linux DVD ISO image of, say, 3.5GB from an intranet site (eliminating the variable of “last mile” Internet connectivity) may experience average throughput of (for the sake of argument) 50Mbit/sec. from an 802.11g access point. At this rate, the download will finish in about (3.5GB*8b/B)/0.05Gbit/sec.=) 560 seconds or somewhat over nine minutes.
Add nine more computers downloading the same image, and the task will be completed in slightly over one and a half hours ((3.5GB*10*8b/B)/0.05Gbit/sec.=5,600 seconds). Those same 10 computers connected to a switch at 100Mbit/sec. with an uplink to the server at 1Gbit/sec. would complete downloading the image in 1/20 the time — or about five minutes. Other variables, such as contention, packet overhead and signal strength are ignored in this example, but it illustrates the magnitude of the bandwidth difference well.
There are inherent limitations to the number of connections to an access point. A good rule of thumb for designing is that each access point can support 20-30 simultaneous users. Applications that don’t demand high bandwidth and/or low latency will suffer performance when the number of simultaneous access point users is too great.
How the access points connect to the backbone is important as well. In the previous example, the wireless access point was assumed to have a wired 100Mbit/sec. connection as an uplink. What about in mesh networks? Depending on how the mesh is laid out, an uplink at 54Mbit/sec. may connect to an access point that feeds two others, and the uplink capacity to the backbone network for each of the two edge access points is therefore decreased by 50%. More access points relying on the uplink would mean a further reduction in bandwidth.
If the primary use is for Internet access, the bandwidth limitation typically lies in the ISP connection. In those cases, LAN bandwidth limitations are usually not much of a consideration. If mobility and cost is, wireless may be the way to go, but there is also the security aspect. Wireless network connectivity adds security issues becau