Obviously, the less overhead, the better your performance will be. However, there are two ways to achieve this efficiency in a packet. The first is to make the headers smaller in an absolute sense. The second is to use larger packets, which makes the overhead less only relative to the data.
To understand the first category, consider that the headers in a typical packet for each protocol, such as the Ethernet, IP, UDP (User Datagram Protocol) and RTP (Real Time Transport Protocol) headers, could be much smaller, but this would come at the expense of flexibility and functionality. For instance, if we didn't spent one byte of every packet on type of service, our networks would be a little roomier, but we would lose the ability to classify packets based on Differentiated Services Code Point or IP Precedence.
One of the most common things people do to improve link efficiency by reducing header sizes is called IP header compression, specified in RFC1144 and RFC2509. You may have noticed this feature on many dial-up network clients. For instance,
A similar method, but one more appropriate for VoIP, is RFC2508, which specifies the Compressed Real Time Protocol (CRTP). This protocol is easy to configure, loses no functionality, and dramatically improves efficiency on a link. It does this by taking the IP header (20 bytes), the UDP header (8 bytes) and the RTP header (12 bytes), which combine to create a 40 byte header, and replacing them with a single header that is often only 2 to 4 bytes. Given that voice traffic is typically a large number of very small packets, using CRTP (assuming your VoIP is using RTP) can easily cut your bandwidth needs in half on PPP links.
As was mentioned earlier, the other method for achieving link efficiency is changing the size of the packets to include more data per header. To explain how this works, let's consider two file transfers where we send a 1,000,000-byte file. For simplicity, we will pretend our transfer program leaves us with a 40-byte header. In the first case, let's use the Ethernet minimum frame size of 64 bytes, and in the second, we'll use the maximum 1518 bytes.
64 bytes total - 40 bytes overhead = 24 bytes of data per packet for the first transfer
1518 bytes total - 40 overhead = 1478 bytes of data per packet for the second transfer
1,000,000 bytes of data / 24 bytes per packet = 41,666 packets in the first transfer
1,000,000 bytes of data / 1478 bytes per packet = 676 packets in the second transfer
41,666 packets x 40 bytes = 1,666,640 bytes of overhead for the first transfer
676 packets x 40 bytes = 27,040 bytes of overhead for the second transfer
Thus, the first transfer takes 2,666,640 bytes and is 62% overhead, while the second transfer puts 1,027,040 bytes on the wire and is 3% overhead.
These basic calculations are the fundamentals behind the second category of networking innovations. The advantage in efficiency for large packets is obvious and has become the norm. However, now we have other applications entering the network that need to minimize delay rather than maximize throughput. So we need a solution that lets us have link efficiency and low delay and jitter. The most common of these solutions is fragmentation and interleaving.
The concept behind fragmentation and interleaving is simple. Because it's impractical to make all the packets the same size, and we can't have large packets delaying our voice packets on slow WAN links, we will break up these big data packets at the link layer and splice the voice packets in between them.
The first thing you should understand about this process is that fragmentation and interleaving is completely local to a data link, unlike IP fragmentation. That is, once a packet is fragmented at the network layer, using the fragmentation bits in the IP header, the packet stays fragmented across all its hops until all the fragments reach the destination, where the receiver is responsible for reassembling the fragments. Instead, in a data-link layer fragmentation, such as Frame Relay's FRF.12, the packet is fragmented as it is placed into a permanent virtual circuit (PVC) and reassembled by the router on the other side of that PVC as it is removed.
Next, fragmentation and interleaving won't do you a lot of good without quality of service. You need to classify the voice packets as higher priority than the fragments of data. Otherwise, a lot of fragments will still be sent ahead of your voice and you'll be worse off than you started.
Last, know that fragmentation and interleaving add a little delay to reduce a lot of jitter. The extra processing time to fragment, interleave and reassemble is fairly constant and, fortunately, is probably not noticeable to users. Jitter, which is the delay to a handful of packets, caused by a long wait while a data packet is clogging up a WAN link, is much more noticeable to users. In other words, if your problem is that all your packets have too much delay, then fragmentation and interleaving is probably not the solution you're looking for.
About the author:
Thomas Alexander Lancaster IV is a consultant and author with over ten years experience in the networking industry, focused on Internet infrastructure.
Read all of Tom Lancaster's #VoIP Tips: http://searchnetworking.techtarget.com/tipsIndex/0,289482,sid7_tax286859,00.html
Visit our extensive collection of #Best Web Links on Voice/data Convergence: http://searchnetworking.techtarget.com/bestWebLinks/0,289521,sid7_tax3a3,00.html
The upcoming Networking Decisions Conference, October 16-18 in Chicago, is focusing on VoIP as one of the most important technologies you need to know about. Featured sessions will include:
The real ROI of VoIP, presented by Steve Leaden, president and founder, Leaden Associates. VoIP no longer stands for just voice over IP, it now includes video over IP and even voice audio conferencing over IP. Steve will discuss how to leverage ROI opportunities in all three areas to facilitate your IT/Telecommunications investment.
The benefits and pitfalls of VoIP, presented by Zeus Kerravala, vice president, enterprise computing, Yankee Group. Zeus will explore the numerous pros and cons of IP telephony and the problems to avoid for the most successful installation.
VoIP: Can you hear me now? presented by Irwin Lazar, practice manager, Burton Group. Irwin will discuss network requirements such as quality of service, performance management, resiliency and security.
Don't miss out! Learn more about all our conference sessions and speakers: http://www.networkingdecisions.com/html/ci_sessions.htm
This was first published in August 2002