How can ATM switch improve Quality of service and data rate just by incorporating them into the core of IP network? What was used in the sore of IP networks before ATM switches and what is used today for even better QoS nad data rate?
It's important to understand that ATM is a technology from the telco world, not from the IP world.
ATM was a technology created by the telcos with the aim of creating a "broadband integrated services digital network". That means a network that could carry both traditional telco services and new higher rate services.
Traditional telco services, particularly voice calls, required guaranteed bandwidth with low latency and jitter. ATM was designed so that it could deliver these services, while also delivering higher rate best-effort or burstable services over the same infrastructure.
To deliver low latency, traffic on ATM networks was split up into small cells. This kept store and forward delays in ATM switches to a minimum. It did, however, come at a fairly substantial cost in bandwidth efficiency.
IP was a technology designed to interconnect a wide variety of different networks. IP was a "best effort" packet switched protocol. "Premium IP" services are certainly possible, but it's not the primary focus of the protocol.
People built IP networks over whatever technology was available, some networks may have been lucky enough to have dedicated lines, but in most cases, long distance links would be leased from telephone companies. Afaict the main practical advantage of ATM is it allowed telcos to sell burstable links to their customers. So rather than buying a fixed rate link and leaving it mostly idle a customer could buy a variable rate link.
As to why ATM went away, I think there were a number of factors.
- The telco industry struggled to bring new technology to market on a fast enough timescale and at a low enough cost for it to remain competitive.
- The "cell switching" concept of ATM made sense when typical data rates were of the order of a few megabits per second, but it makes far less sense in a multigigabit world. At 10 megabits per second a 1500 byte packet takes 1.2 miliseconds to serialise, at 10 gigabits per second it takes only 1.2 microseconds.
- WDM and optical amplification became available, this allowed a single fiber to carry multiple full-rate data streams.
- Many countries introduced "local loop unbundling" allowing ISP's direct access to customers phone lines, rather than being forced to backhaul data over the incumbent telco's network they could build their own networks.
- "Best effort" internet traffic grew at a rate that massively outpaced "premium" voice traffic. Premium services at high data rates never took off at least in the consumer market.
Ethernet started out as a Local area network technology, the CSMA/CD technology was fundamentally distance limited. However when the introduction of full duplex mode that limit went away. Ethernet was fast, cheap and expertise in running it was readily available, so it was an attractive technology for new network operators. Over the last couple of decades it has taken over the networking world.
Pure Ethernet does not scale very well to large networks, so in larger networks it is common to simply treat the Ethernet connections as point to point links. Depending on the needs of a particular network, IP may be run directly on top of Ethernet or an intermediate layer like MPLS may be used.