ATM is just a technology. Over the last decade, depending on who had the faster chipset, folks have chosen an ATM chipset, an Ethernet or even a Frame-Relay chipset for their NAPs/Exchange Points. At one point in time, the Gigaswitch was the fastest available. Then ATM switches could do 155Mbps and OC12. GigEthernet and 10 GigEthernet appeared, and folks went the other way. Both ATM and Ethernet exchange points have route-servers. Equinix, PAIX, SBC, and Worldcom have route-servers at their exchange points in the USA. LINX and other have route-servers at exchange points outside the US. Fully filtering your peering sessions with route-servers is relatively simple. ATM has PVCs, Ethernet has VLANs. Ethernet interface cards tend to be less inexpensive than ATM interface cards. ATM is a natural wide-area technology, Ethernet is a natural local-area technology. Frame-relay tends to be used for smaller exchange points, although it could also work in a mid-sized exchange points. The reliability between ATM and Ethernet isn't that different over the long term. Most of the problems tend to be with the end-connections (congestion and broadcast loops), and not the fabric itself. But there have been notable outages with atm switches and ethernet switches at exchange points. Multicast is an interesting problem in itself. None of the ATM switches handle multicast very well, and even the ethernet switches have problems. On Mon, 18 December 2000, Jay Ford wrote:
On Mon, 18 Dec 2000, Ben Buxton wrote:
I'm interested in peoples thoughts as to what is happening with the future of ATM NAPs. Are people moving away from them or are they still quite popular? Downsides or upsides? I'm trying to gauge whether they are worthy of investing equipment and hassle into for peering...
From a research university perspective, the AADS NAP is a cool thing. It lets me do peering & various types of transit on a single circuit, & the availability of the route servers is nice. The full mesh of PVCs removes most of the layer 1-2 pain involved with firing up new interactions. The circuit to get there isn't cheap, but it seems worth it based on my experience.