An interview in which Silver Peak claimed to be able to accelerate UDP drew something of a counterblast from Riverbed. Among other things, it commented that, "Virtually all enterprise traffic runs on TCP, which is why companies such as Riverbed focus on TCP", and "Although 40 percent of traffic is UDP, you can't accelerate it so its relevance in the market is reduced."
But is that true? I put the point back to Silver Peak's product marketing director Jeff Aaron. He contends that most WAN optimisers operate at the TCP layer, so they do indeed have problems accelerating UDP, but that his company's devices can do it because they operate at the network layer. That means that when it fingerprints data blocks to detect and remove repetition (ie. when it does data reduction), it can do so at finer granularity, he claims.
At present, much of the UDP traffic crossing enterprise WANs will be VoIP, and if there's one thing we can say with certainty, it's that VoIP traffic is going to grow and grow. But VoIP cannot be accelerated, according to Riverbed's EMEA marketing director Mark Lewis.
"VoIP is a special category of network traffic that requires separate and distinct consideration from the more traditional TCP-based traffic found in IP networks," he says. He adds that users have much higher expectations for telephony services for web and file applications, so VoIP traffic must be handled with care to avoided degraded voice quality and dropped calls.
"VoIP is considered real-time traffic; VoIP datagrams must be delivered by the network with a minimal amount of jitter in order to maintain voice quality," Lewis continues. "Latency must be kept below 150ms if possible. Any deviations from these standards will affect voice quality. For these reasons, any techniques employed to expedite delivery of VoIP traffic must not add latency or jitter.
"To 'accelerate' VoIP traffic is an ambiguous, often-misleading phrase. VoIP traffic moves at the speed of light, just like all other types of IP traffic. It can be delayed by network congestion and various other issues, resulting in degraded voice quality and dropped calls. But you cannot make VoIP packets travel faster than the speed of light, and you cannot make the participants talk any faster."
Indeed, Lewis only highlights one area in which WAN optimisation can help VoIP, and that is bandwidth management. "If adequate bandwidth does not exist, then VoIP quality will suffer," he says.
Jeff Aaron agrees that there are specific factors preventing you from speeding up VoIP traffic, but says the ability to accelerate UDP can still bring benefits to VoIP users.
"90 percent of our tenders today are for companies who either have VoIP or are planning to have it," he notes.
"Whether you can accelerate voice is a semantic issue. You can't make it go faster but you can improve the quality and performance, do loss mitigation, add quality of service, improve its efficiency, add forward error correction and even do data reduction." All those factors can translate into more calls on the same bandwidth with fewer dropped packets, he says.
And he adds that the ability to accelerate UDP is not just relevant to VoIP anyway, and can only get more important in the future: "We're seeing enormous interest in disaster recovery, and in particular in data replication. Where that's unique is that UDP is the default transfer mechanism for many DR applications," he says. "DR is 90 percent repetitive, so fingerprinting and data reduction can give a big improvement."