Home >> May 2012 Edition >> Tech OPS: XipLink Real-Time Optimizations
Tech OPS: XipLink Real-Time Optimizations
by Charlie Younghusband, Founder and Senior Vice President of Product Management, XipLink

XipLink Real-time (“XRT”) is a new optimization capability that compresses, coalesces and prioritizes VOIP and UDP for significantly more bandwidth and packet efficiency without compromising quality. XRT can provide bandwidth savings up to 50 percent and guarantee quality delivery. This article describes how small packet applications can benefit from optimization, how XRT functions, how XRT can be added to a network to deliver large bandwidth and packets per second reductions, and then specific XRT savings examples.

Vice XipLink Real-time (“XRT”) is a new optimization capability that compresses, coalesces and prioritizes VOIP and UDP for significantly more bandwidth and packet efficiency without compromising quality. It provides key benefits including dramatic bandwidth reduction, significant reduction in the packets per second carried over a network, and ensures that real-time traffic is properly prioritized for optimum quality.

XRT optimizes small packets, in particular VoIP. However, Skype and other UDP/TCP applications with many small packets can also benefit from XRT. While there are now many highly bandwidth efficient VOIP codecs available, they cannot address the header overhead. XRT removes the high overhead associated with IP packet delivery. XRT is available through XipLink’s flexible and scalable line of Appliances (XA).

The amount of voice traffic on satellite links continues to increase at a rapid pace, with significant growth in the use of VoIP, Skype and other voice applications. Real-time traffic such as voice, which is based on small UDP packets, places a tremendous load on network devices due to the high packets per second per voice call. In addition, voice traffic such as H.323 based VoIP or Skype, is highly bandwidth inefficient as the packet headers are often the same size as the payload—an overhead rate of 40 percent.

How can VOIP and UDP traffic benefit from optimization? It’s true that they cannot be ‘accelerated’ in the same way that TCP traffic can be optimized, but there are important other optimizations that help both the quality, timeliness and bandwidth use of these real-time applications.

VOIP and other streaming application can benefit from optimization in many ways:

xiplinkfig1 Approximately 40 percent of most VOIP packets are headers.

Most headers do not change much from packet to packet: there is a lot of redundancy between packets—even as they are sent milliseconds apart.

This redundancy is true even with multiple streams (calls): calls may often have a common destination or source IP address, but there is no sharing of headers between them.

Inefficiency applies to all UDP—standard VOIP as well as Skype etc., but in particular small packet applications (e.g. not full frame video streaming).

These real-time steams need to be prioritized but managed through the network.

Effective voice codecs are well established and provide excellent options to tune quality versus size. Little need to touch the data stream.

Jitter and delay are critical in VOIP, but optimization is possible with minimal impact and in many cases will benefit from quality improvement at the same time.

All of these issues and weaknesses are addressed by XRT.

XRT Capabilities
The primary features of the XipLink Real-time include:

Header Compression—ROHC Header Compression
Coalescing—Concatenate small packets together while minimizing jitter
QoS—Shaping and bandwidth control of traffic

Header Compression
The XipLink solution incorporates an implementation of the Robust Header Compression (ROHC) standard. This is an IETF standard and the most modern and effective technique for header compression. Header compression takes advantage of the streamed nature of the UDP protocol, since most headers stay the same or are similar (IP address is always the same, etc.), a small reference token is substituted for the repetitive fields. This typically reduces the packets from 40 bytes to 6 bytes or less. The XRT solution has profiles to compress the IP headers, UDP, UDPlite and then RTP for VOIP. So a UDP based application, like Skype, can have its UDP and IP headers compressed while maintaining it’s own proprietary (non-RTP) session header.

While the largest benefits can be seen from VOIP, an IP application can also benefit from XRT if it has streams of small packets, including TCP applications such as Citrix.

ROHC is substantially more effective than older compression techniques such as cRTP, for compression effectiveness, and for resilience against packet loss.

“Coalescing” groups multiple header-compressed packets together into a single packet. This can also be termed as aggregation. Compressed packets are concatenated within the coalesced packet. This allows the IP headers of individual packets to be compressed, yet provides an ultra-lightweight shell that provides route-ability and the correct DSCP mark as the packets it contains.

Multiple coalescing queues are used, one for each DSCP class. This ensures that the QoS is maintained throughout the network. At the most basic level, it ensures that high priority traffic is not compromised by coalescing a larger non-priority packet, such as coalescing VOIP together with video. It also allows for multiple priority queues to be active at the same time, for instance it automatically supports Multi-Layer Protocol Prioritization (MLPP) in U.S. government and military applications.

Once the capture window timer expires or the coalesced maximum size is reached, the coalesced packet is sent immediately. The capture window and maximum size of the coalesced packet can be configured. On a low bandwidth network, a smaller maximum packet size can be configured so the serialization delay does not contribute to jitter. A longer capture window can provide more benefit if there are only a few calls on a network link, but for VOIP, the value must be safely less than the jitter buffers of the phone systems. No packets are sent if there is nothing to coalesce.

xiplinkfig3 Despite the header compression, XRT has a valid IP header address so it is fully routable through any network.

XRT is completely independent of the VOIP codec or payload being used, and different codecs can be combined in the same coalesced packet. To ensure that the QoS is properly enforced on the packet as it continues to traverse the network, a different coalescing queue is used for each DSCP value. This also helps ensure that a large video does not get lumped in with a more latency sensitive VOIP packet.

The XA Appliance User Interface (UI) provides a means to configure what traffic should be compressed. VOIP traffic must be configured explicitly to use the RTP ROHC profile for maximum compression benefit, usually by specifying the call manager IP address on the network.

QoS + Prioritization
XipLink Appliances also incorporate advanced QoS functionality, which is tightly coupled with other XipLink optimization functions. The QoS is also capable of working harmoniously with networks that already have an existing QoS implementation to support voice based applications.

On networks with an existing QoS system, XRT ensures that optimized traffic retains the same DSCP tags, such that the QoS through the network is maintained and the VOIP is properly prioritized and served.

On networks without QoS, as is typical of SCPC technologies, the XipOS QoS provides operators with the ability to prioritize and shape traffic using a Hierarchical Class Based Queuing technology. In addition, the operator can configure real-time, maximum and relative priorities. It also provides features specifically of interest for real-time traffic delivery, including latency bounded queues which is a highly desirable feature for managing real-time traffic effectively. These features give an operator a very rich classification engine, not only intended to classify the voice packets, but to simultaneously prioritize and ensure quality SLAs are achieved.

In both cases, the end result will be bandwidth efficient, high quality voice and properly prioritized delivery of real-time data.

Let us examine the most popular codec for bandwidth efficient networks, the G.729A codec. The following diagram illustrates that the benefits associated with XRT delivers approximately a 50 percent savings in bandwidth after just five calls on any particular link.

xiplinkfig4 When testing using an Ixia IXLoad device over an emulated satellite link, the benefits can be clearly seen. Importantly, the MOS quality score that is auto-calculated by the Ixia device remained at 3.4, both with and without optimization. A MOS score of 3.4 is essentially inherent with the G.729 codec with a satellite delay. The use of QoS protects call quality and MOS score against degradation when the network congests.

The benefit varies with the payload; smaller payloads with a higher percentage of header-to-payload will benefit the most. Let us examine the benefit across a sample of different codecs that may be used, with an assumed volume of 10 calls.

It can be seen that very high data rate VOIP codecs such as G.711 stand to benefit the least, as they carry a large payload compared to the header.

Protocols such as Skype or any other UDP streaming application do not benefit from the RTP level compression, but do benefit from having the UDP and IP headers compressed. A typical 28 percent bandwidth savings for Skype is still very significant.

Predicting XRT Benefits
XipLink has a calculator tool that models the compression solution and accurately predicts the benefit that can be expected from XRT, including the bandwidth and the PPS, using packet sizes, traffic type and packet arrival rates as inputs. Configuration options available from the XipLink UI including the capture window and maximum coalesced packet size can be configured. The XipLink calculator tool is available on request.

The XRT solution is a feature of all XipLink XA Appliances and is included free of charge with all XA units. XRT runs simultaneously with other XipLink optimization techniques. The solution can be deployed over point-to-point links as well as hub-spoke networks common to TDMA and meshed network architectures. XA Appliances can be installed inline, use policy based routing, or WCCP for out-of-path deployment; they just simply need to be installed so they can intercept the traffic requiring optimization.

The solution makes use of XipLink Lightweight Tunnels (XLT). XLT has only a one byte header on top of UDP/IP so it is highly bandwidth efficient. At the hub of a network, the XA appliances receive tunnels from the remotes. Up to 1,000 tunnels can be supported simultaneously.

Remote units are simply configured with the IP address of the tunnel server and a password. If the tunnel cannot be established, packets are sent without optimization.

xiplinkfig5 The XRT solution currently does not support the optimization of VOIP between nodes in a meshed network. However, packets going remote-to-remote will work automatically but will not see any compression benefit. If there is traffic to a common node, which is common of meshed networks, those links can be fully optimized.

Consult the model specifications to see how many calls and PPS are supported per particular model.

Bandwidth Savings + Easy Deployment
This article examined how using XRT on a network can provide large bandwidth savings for small packet applications and ensure high quality packet delivery through the network. It can be deployed easily. XRT can be installed with the primary purpose of saving VOIP bandwidth or as part of an integrated solution to bandwidth optimization with XipLink technology.

Packet coalescing capability and QoS capabilities are available in XipOS version 3.2. Header compression capability is available in XipOS version 3.3.