Ethernet's metro traffic police
more on the topic
One start-up aims to protect SLAs and boost network usage
Loss is a Fact of Life in packet-based networks. Traffic often progresses easily through uncongested access networks, only to get deleted in an aggregation point already clogged with more important, time-sensitive traffic — such as video conferencing. And that's logical, but as the folks at a new metro Ethernet equipment vendor will point out, that traffic wasted a lot of bandwidth along the way just to be deleted in the end.
The philosophy of Ethos Networks, an Israeli vendor exiting stealth mode this year, is that if the network is so congested in some places that some traffic will be deleted, then that excess traffic shouldn't be allowed in the network in the first place, so that the bandwidth it might consume can be put to better use.
Ethos' switch, available in June, doesn't even let traffic into the metro network unless it's certain that traffic can reach its destination without violating the service level agreements (SLAs) attached to it. Once capacity is secured for services with committed information rates, the remaining capacity is allotted evenly among lower-tier or best-effort services. The result is not only SLA assurance but better network usage.
“If you know in advance the fate of every packet before it goes into the network, then you can give concrete guarantees to everyone,” said Adam Dunsky, co-founder of Ethos. “Our policy is to never delete packets inside the network, only to delete them at the entry.”
To have this kind of end-to-end visibility of network resources, all the nodes in the network must be in frequent communication, a requirement you'd expect would limit the scale of the system. “I'm trying to create a state where every switch knows exactly what every other switch is doing,” Dunsky said.
But Ethos makes that communication as efficient as possible, minimizing the amount of information those nodes need to share to do their work. For example, rather than constantly updating one another about all activity throughout the network, the nodes only need to communicate about potential traffic collisions and bottlenecks. The nodes can calculate what's happening at any moment based on what was happening moments earlier and where things were headed. And rather than describe everything that's happening, they can just describe what's changed since the last message.
Another factor keeping internode conversation low: Aggregated traffic flow in metro networks, where Ethos' gear is meant to be deployed, is less bursty and more predictable than individual traffic flows. So there's just not as much for the nodes to talk about as you might think. “The result is we can safely hold a domain of 200,000 services and still maintain real-time admission control at the packet level for all of them,” Dunsky said.
Maximizing network usage frees up more capacity. Dividing that extra capacity evenly among lower-tier services should not only make the performance of those lower-tier services more consistent, it should make them, on average, faster and better overall, Ethos maintains.
“A customer buys 100 Mb/s [service],” Dunsky said. “It's best-effort. In effect, he's getting 20 Mb/s. With Ethos, he's getting 60 Mb/s. Or he's getting 60 Mb/s today, but his video is always being interrupted. With Ethos, he'll get 40 Mb/s, but his video will never be interrupted.”
U.S. PUBLIC ETHERNET SERVICE REVENUE
(in billions of U.S. dollars)
2007 $1.433
2008 $2.049
2009 $2.766
2010 $3.569
2011 $4.354
2012 $5.050
Source: Insight Research
popular articles
Want to use this article? Click here for options!
© 2008 Penton Media Inc.












