Continuing my series as I work through the CCNA syllabus. The introduction to the series can be found here.
I will be pretty much following the CCNA Composite Exam Blueprint point for point. One post per bullet point. I’m using Version 11 (640-802).
Describe the impact of applications such as Voice Over IP on a network
This is another tricksy entry. what does that “such as” encompass? Well, it could be pretty much anything, couldn’t it? A better way to view this is to work out how any application impacts the network, and then refine that for different types of traffic.
At root, the network is a transportation system. What we are really interested in is getting something (data) from one point in the system, to another. So we are concerned with two things:
- How much can we transport?
- How long will it take? (and is that time consistent)
Unlike with a lorry, or a boat, or a train, how much to us means how much per second. The bandwidth of the links. We can keep stuffing data into the link until the cows come home, and we probably will. Likewise, the “how long” part is better written as “how long does it take one packet to traverse the network, and get another one back” this is the latency of the link. Jitter is the term used to define how much the latency changes over time. If the first packet takes 2ms to traverse the link, but the second and third take 200, then the fourth takes 5 etc etc, you have high jitter. Most applications work best with low jitter, but some, such as web browsing, don’t require it.
When we progress later to the Internetwork, the biggest factor in these two things will be the route the packets take, and that is why so much focus is put on routing. But in a small simple network, where we like to start, we look at the individual links.
With two constraints (bandwidth and latency) we can easily see that there are three broad classes of applications:
- Requires high bandwidth, doesn’t care about latency.
- Requires moderate latency and moderate bandwidth.
- Requires low latency and doesn’t really care about bandwidth.
Of course, most applications fall between the extremes, but the four most common situations can be fairly clearly defined. Most applications don’t require either a lot of bandwidth, or a particularly low latency. Web browsing is a good example of this. Pages tend to be small compared with a modern LANs bandwidth, and people don’t tend to notice if a page takes a few tenths of a second to load.
File downloads tend towards #1 in our list. Thanks to TCP and it’s windowing, the server can blat out lots of large packets, only requiring responses irregularly, and hence latency is not an issue, but the bandwidth will tend to be used up, no matter how much you throw at it.
Audio (VOIP) requires a very low latency (and also a fairly consistent jitter) but very little bandwidth. Audio will struggle if packets are lost, or out of order, but thanks to compression, will use only a few 10s of kbps of bandwidth. Video, is a bit of a hybrid. It requires the same latency and jitter as audio (indeed, it probably uses the same audio codec, so has the same constraints!) but also requires a higher bandwidth for the video part. Again, thanks to modern video codecs, the bandwidth required can be kept relatively low.
The final application, is the only one that has any real reliance on jitter, and that’s gaming. As the game engines need constant updates to keep each machine in sync, a small amount of information is sent very often, and a high jitter interrupts this process, leading to poor performance. The vast majority of networks consider gaming to be the lowest of low priorities though, so very little consideration of this is put into network design.
So, to get back to our question. What is the impact of Voice over IP on a network? The fact that we are carrying voice traffic means that we need to be concerned with the latency of the link. What adds latency? There are three main items, all interconnected. The first is the length of cable traversed. In both optical fibre, and UTP the signal travels at the speed of light. That signal can circumnavigate the globe 7 times in a second. But if our traffic is degraded when the latency gets above a few 10s of ms, and it is travelling from Europe to the US, the travel time can become noticeable. This becomes even more so when Satellite signals become involved, as they are far enough from the earth that the signals can take 10ths of seconds to travel to the satellite and back to the ground station.
The second big impact to latency is the devices in the path. A switch takes a small time (usually measured in ns) to process a packet, strip off the headers, insert new ones, and forward out of a port. A router takes even longer, and even longer still if ACLs become involved and the packet has to be shifted from ASICs into the main CPU of the router. Interestingly, hubs do not add processing time, as they are simple electrical connections, they act as an imperceptible bump in the cable. The final part is the switch from optical to electrical signal in fibre. When the light is processed to become an electrical signal, it takes a very small time to process the signal. This, like a hub, is generally not noticeable, but if a lot of fibre hops are involved, will add up.
So, how do we design a network to accommodate low latency traffic such as VOIP? Firstly, we keep cable runs as short as possible. The star topology helps with this, as devices are cabled to a “central” point, rather than in a ring that gets longer as more devices are added, and that must circumnavigate the campus. The second way is to keep the number of switches low. Adding switches, adds latency, so we design with fewer, higher density switches. Finally, we do the same with routers. Fewer routers, leads to less processing time, and lower latency. Also, keeping routers close to the centre of the network, and firewalls to the edge of the network, means that much VOIP traffic traverses less hops, and stays in the fast switching circuits when it does. Only if it is leaving the network is it dropped back to the CPU to be inspected.
The final option we have, which is not of too much concern for CCNA, but which comes up in the real world is QoS, or Quality of Service. This allows us to tag traffic and have the router handle the tagged traffic first, so that when the network is busy, the higher priority traffic is processed first.
So, how much impact will VOIP have on the network? That depends on the network, and on the amount of VOIP traffic. It may be that the amount is low enough that it doesn’t impact the overall design. Or it may be that extra money is spent to run fibre in order to reduce the amount of switches necessary to “reduce the radius” of the network. Every network is different, but the guidelines above stand:
- Less switches.
- Less routers
- Shorter cable runs
- Less latency.