NetEqualizer Field Guide to Network Capacity Planning


I recently reviewed an article that covered bandwidth allocations for various Internet applications. Although the information was accurate, it was very high level and did not cover the many variances that affect bandwidth consumption. Below, I’ll break many of these variances down, discussing not only how much bandwidth different applications consume, but the ranges of bandwidth consumption, including ping times and gaming, as well as how our own network optimization technology measures bandwidth consumption.

E-mail

Some bandwidth planning guides make simple assumptions and provide a single number for E-mail capacity planning, oftentimes overstating the average consumption. However, this usually doesn’t provide an accurate assessment. Let’s consider a couple of different types of E-mail.

E-mail — Text

Most E-mail text messages are at most a paragraph or two of text. On the scale of bandwidth consumption, this is negligible.

However, it is important to note that when we talk about the bandwidth consumption of different kinds of applications, there is an element of time to consider — How long will this application be running for? So, for example, you might send two kilobytes of E-mail over a link and it may roll out at the rate of one megabit. A 300-word, text-only E-mail can and will consume one megabit of bandwidth. The catch is that it generally lasts just a fraction of second at this rate. So, how would you capacity plan for heavy sustained E-mail usage on your network?

When computing bandwidth rates for classification with a commercial bandwidth controller such as a NetEqualizer, the industry practice is to average the bandwidth consumption for several seconds, and then calculate the rate in units of kilobytes per second (Kbs).

For example, when a two kilobyte file (a very small E-mail, for example) is sent over a link for a fraction of a second, you could say that this E-mail consumed two megabits of bandwidth. For the capacity planner, this would be a little misleading since the duration of the transaction was so short. If you take this transaction average over a couple of seconds, the transfer rate would be just one kbs, which for practical purposes, is equivalent to zero.

E-mail with Picture Attachments

A normal text E-mail of a few thousand bytes can quickly become 10 megabits of data with a few picture attachments. Although it may not look all the big on your screen, this type of E-mail can suck up some serious bandwidth when being transmitted. In fact, left unmolested, this type of transfer will take as much bandwidth as is available in transit. On a T1 circuit, a 10-megabit E-mail attachment may bring the line to a standstill for as long as six seconds or more. If you were talking on a Skype call while somebody at the same time shoots a picture E-mail to a friend, your Skype call is most likely going to break up for five seconds or so. It is for this reason that many network operators on shared networks deploy some form of bandwidth contorl or QoS as most would agree an E-mail attachment should not take priority over a live phone call.

E-mail with PDf Attachment

As a rule, PDF files are not as large as picture attachments when it comes to E-mail traffic. An average PDF file runs in the range of 200 thousand bytes whereas today’s higher resolution digital cameras create pictures of a few million bytes, or roughly 10 times larger. On a T1 circuit, the average bandwidth of the PDF file over a few seconds will be around 100kbs, which leaves plenty of room for other activities. The exception would be the 20-page manual which would be crashing your entire T1 for a few seconds just as the large picture attachments referred to above would do.

Gaming/World of Warcraft

There are quite a few blogs that talk about how well World of Warcraft runs on DSL, cable, etc., but most are missing the point about this game and games in general and their actual bandwidth requirements. Most gamers know that ping times are important, but what exactly is the correlation between network speed and ping time?

The problem with just measuring speed is that most speed tests start a stream of packets from a server of some kind to your home computer, perhaps a 20-megabit test file. The test starts (and a timer is started) and the file is sent. When the last byte arrives, a timer is stopped. The amount of data sent over the elapsed seconds yields the speed of the link. So far so good, but a fast speed in this type of test does not mean you have a fast ping time. Here is why.

Most people know that if you are talking to an astronaut on the moon there is a delay of several seconds with each transmission. So, even though the speed of the link is the speed of light for practical purposes, the data arrives several seconds later. Well, the same is true for the Internet. The data may be arriving at a rate of 10 megabits, but the time it takes in transit could be as high as 1 second. Hence, your ping time (your mouse click to fire your gun) does not show up at the controlling server until a full second has elapsed. In a quick draw gun battle, this could be fatal.

So, what affects ping times?

The most common cause would be a saturated network. This is when your network transmission rates of all data on your Internet link exceed the links rated capacity. Some links like a T1 just start dropping packets when full as there is no orderly line to send out waiting packets. In many cases, data that arrive to go out of your router when the link is filled just get tossed. This would be like killing off excess people waiting at a ticket window or something. Not very pleasant.

If your router is smart, it will try to buffer the excess packets and they will arrive late. Also, if the only thing running on your network is World of Warcraft, you can actually get by with 120kbs in many cases since the amount of data actually sent of over the network is not that large. Again, the ping time is more important and a 120kbs link unencumbered should have ping times faster than a human reflex.

There may also be some inherent delay in your Internet link beyond your control. For example, all satellite links, no matter how fast the data speed, have a minimum delay of around 300 milliseconds. Most urban operators do not need to use satellite links, but they all have some delay. Network delay will vary depending on the equipment your provider has in their network, and also how and where they connect up to other providers as well as the amount of hops your data will take. To test your current ping time, you can run a ping command from a standard Windows machine

Citrix

Applications vary widely in the amount of bandwidth consumed. Most mission critical applications using Citrix are fairly lightweight.

YouTube Video — Standard Video

A sustained YouTube video will consume about 500kbs on average over the video’s 10-minute duration. Most video players try to store the video up locally as fast as they can take it. This is important to know because if you are sizing a T1 to be shared by voice phones, theoretically,  if a user was watching a YouTube video, you would have 1 -megabit left over for the voice traffic. Right? Well, in reality, your video player will most likely take the full T1, or close to it, if it can while buffering YouTube.

YouTube — HD Video

On average, YouTube HD consumes close to 1 megabit.

See these other Youtube articles for more specifics about YouTube consumption

Netflix – Movies On Demand

Netflix is moving aggressively to a model where customers download movies over the Internet, versus having a DVD sent to them in the mail.  In a recent study, it was shown that 20% of bandwidth usage during peak in the U.S. is due to Netflix downloads. An average a two hour movie takes about 1.8 gigabits, if you want high-definition movies then its about 3 gigabits for two hours.   Other estimates are as high as 3-5 gigabits per movie.

On a T1 circuit, the average bandwidth of a high-definition Netflix movie (conversatively 3 gigabits/2 hours) over one second will be around 400kbs, which consumes more than 25% of the total circuit.

Skype/VoIP Calls

The amount of bandwidth you need to plan for a VoIP network is a hot topic. The bottom line is that VoIP calls range from 8kbs to 64kbs. Normally, the higher the quality the transmission, the higher the bit rate. For example, at 64kbs you can also transmit with the quality that one might experience on an older style AM radio. At 8kbs, you can understand a voice if the speaker is clear and pronunciates  their words clearly.  However, it is not likely you could understand somebody speaking quickly or slurring their words slightly.

Real-Time Music, Streaming Audio and Internet Radio

Streaming audio ranges from about 64kbs to 128kbs for higher fidelity.

File Transfer Protocol (FTP)/Microsoft Servicepack Downloads

Updates such as Microsoft service packs use file transfer protocol. Generally, this protocol will use as much bandwidth as it can find. There are several limiting factors for the actual speed an FTP will attain, though.

  1. The speed of your link — If the factors below (2 and 3) do not come into effect, an FTP transfer will take your entire link and crowd out VoIP calls and video.
  2. The speed of the senders server — There is no guarantee that the  sending serving is able to deliver data at the speed of your high speed link. Back in the days of dial-up 28.8kbs modems, this was never a factor. But, with some home internet links approaching 10 megabits, don’t be surprised if the sending server cannot keep up. During peak times, the sending server may be processing many requests at one time, and hence, even though it’s coming from a commercial site, it could actually be slower than your home network.
  3. The speed of the local receiving machine — Yes, even the computer you are receiving the file on has an upper limit. If you are on a high speed university network, the line speed of the network can easily exceed your computers ability to take up data.

While every network will ultimately be different, this field guide should provide you with an idea of the bandwidth demands your network will experience. After all, it’s much better to plan ahead rather than risking a bandwidth overload that causes your entire network to come to a hault.

Related Article a must read for anybody upgrading their Internet Pipe is our article on Contention Ratios

Created by APconnections, the NetEqualizer is a plug-and-play bandwidth control and WAN/Internet optimization appliance that is flexible and scalable. When the network is congested, NetEqualizer’s unique “behavior shaping” technology dynamically and automatically gives priority to latency sensitive applications, such as VoIP and email. Click here for a full price list.

Other products that classify bandwidth

2 Responses to “NetEqualizer Field Guide to Network Capacity Planning”

  1. The Real Killer Apps: What You Can Do to Stop Them from Bringing Down Your Internet Link « NetEqualizer News Blog Says:

    […] For a  more detailed discussion on how much Bandwidth specific applications consume, click here. […]

  2. Charles Everett Says:

    That was a good article but its more helpful if it adds some information regarding recommendations on testing.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: