Failed roll-outs are a thing of the past, claims Itheon, which has announced a new emulator that it claims can replicate WAN conditions such as latency, bandwidth and packet loss - thus allowing proper testing to be done in the lab.

The Itheon Network Emulator is a Linux-based portable PC with software that emulates a network by injecting latency and adjusting both the bandwidth and the error rate. It can create multiple virtual networks too, for example to show the layered effects of a LAN and a WAN.

"A lab can easily test bandwidth, but almost no-one puts real latency in. People say they calculate the effect of latency but that's bollocks - you can't do that. Not even the application vendor knows enough about it to calculate latency," says Itheon networking specialist Cliff Chapman.

He says he's seen plenty of projects fail due to WAN latency: "People forget that a transaction could be dozens of internal calls, each one of which gets slowed down, so all of a sudden it's not 100ms but eight seconds, because they all stack up."

The need to work over WANs has spawned a whole sub-industry, with companies such as Expand, Fineground, Packeteer, Peribit and Riverbed all coming up with different ways to accelerate application performance. Chapman warns though that techniques such as data compression do nothing to improve network latency.

"You can always give an application more bandwidth, but if your problem is latency, no amount of bandwidth will fix that," he says, noting that latency can hit throughput too, by causing TCP to throttle back its transmit rate. He adds that INE can also simulate the effects of routed QoS, and says that at around £8,000, it is less than half the cost of alternatives such as Shunra's Storm.

However, Sunra's UK sales director Daniel Denham says that tools like INE only cover single chunks of the application lifecycle. He claims that Storm is the only product to go all the way from programming through application QA to WAN testing. Denham argues it is vital to support the complete lifecycle, pointing out that many developers - for example games publishers - only have one chance to get it right, as they cannot afford product recalls.

Itheon reckons there is an especial need for pre-deployment testing though. "People say 'I'm going to roll out Oracle, how much bandwidth do I need per user?' But that's the wrong question," says Chapman. "It should be 'What's the performance going to be, given 10k per user or 20k per user?' We can test that. You could also use it to compare the effects of using Peribit, Riverbed or Packeteer, for example."

To set up INE, you tell it what WAN characteristics to apply to through traffic. Chapman admits that real networks will always have extra vagaries of their own, but argues that any testing is far better than none. He adds that a future version of INE will emulate specific real WANs by monitoring a connection, pulling in its characteristics, and then replaying those network conditions on demand.