Foundry Networks has sparked a storm of controversy by claiming its Gigabit Big Iron 8000 range can offer true 'ATM like' Quality of Service (QoS) in an Ethernet environment.
Citing testing data from the NSS Group, the Gigabit vendor said it can make connectionless Lans - which don't allow networks to plan for traffic or refuse to support a relationship that exceeds resources - deliver QoS.
"NSS has confirmed what the IT industry has suspected for a long time," said Bobby Johnson, president and CEO of Foundry Networks.
"ATM has sold itself for too long on being the only networking technology that offers true QoS, but the NSS test demonstrates that this simply isn't the case.
"Big Iron can support massive amounts of mixed traffic via four levels of prioritisation, which allows mission critical applications, including video and voice, to always be available. Quality is not compromised."
However Mark Iliffe, European sales support manager at Foundry, was more pragmatic.
"There is fact and fiction. The idea of the NSS test is to open doors for our salesmen to new accounts, which may have only considered ATM capable of delivering QoS in the past. Unless you're an ATM bigot, then the test demonstrates that Gigabit Ethernet should be treated as a serious alternative."
Sorting out your priorities
The test details show that Foundry is equating Layer 4 switching and switch prioritisation with QoS. However, prioritisation is a characteristic of a network element, while true ATM like QoS is a network wide attribute - so Johnson appears to be mixing one with the other.
The key is that true QoS works across an entire network, not just a single element like a switch. A switch router, like the Big Iron 8000, only provides traffic prioritisation which is just one factor involved in providing end to end QoS.
Iliffe admitted that many aspects of ATM like QoS were still some way off. "It would be great if we could run Diffserv across the network, to provide class of service (CoS) or QoS end to end," he said.
"Over time, we will add QoS functionality and capabilities to our products. Providing directory services is in the product plan, as well as providing new modules, which will address the need to look even deeper into packets."
All these strategies require that the network be able to identify flows that are to receive special QoS treatment. This should be possible using something like the combination of originator and destination network address (IP address, for example) and socket or transport address.
Any strategy that aims to provide a level or grade of service requires that users provide a rough measure of the traffic to be presented - and often some characteristics about the way the peak-and-valley relationships in the traffic will fall.
Iliffe admitted: "There are a lot of things that need to happen to fully implement end to end QoS services over switched Ethernet. There are no application-aware protocols out there, and today, no application can make a QoS request to the network."
Why ATM doesn't rule the world
Defending Foundry's switch architecture, Iliffe said: "The standards bodies don't think more than eight priorities of traffic need to be supported, and most manufacturers of Ethernet switches only support four levels of prioritisation."
However, while the IETF may define eight traffic priorities, it does not mandate how vendors design their switches. There is nothing to stop a vendor supporting, say, 200 discrete queues per switch port for eight traffic priorities, except the expense involved.
As most applications don't have any QoS at all, is it realistic to assume that these applications will suddenly inherit a requirement for extreme QoS sensitivity?
Also, is it realistic to assume that new applications, not even run today, will be so business-critical that a complete network upgrade will be needed to support their QoS demands?
The answers to these questions is almost certainly not, which is why ATM doesn't rule the world. It's probably the same reason we won't see many QoS enabled, directory enabled, policy managed Ethernet networks implemented. Even if the standards and interoperability issues are finally settled.
Delivering Quality of Service
There are several options which deliver QoS:
- Simple over-provisioning.
Moving to 100BaseT or Gigabit Ethernet would increase network resources, and thus improve service quality.
- Prioritisation of specified flows.
The way resources are allocated to traffic can be impacted if the network can be made to recognise priority flows and grant them preferred access to resources. This changes the balance of resources, but doesn’t necessarily guarantee that the end user gets what they’re requesting.
- Mapping of connectionless flows to a connection oriented subnetwork.
This is the ATM model, where a Lan virtual structure is built (via LANE or MPOA) onto an ATM connection network. Because the ATM network can provide per-connection (and thus per-flow) resource management, it's possible to more directly control QoS.
- Creating a connection-oriented subset of the connectionless network.
This is what is done with RSVP, 802.1p, Intserv, Diffserv, etc. The network must identify a priority flow, but when it does that it uses an end to end signalling procedure like a connection oriented call set-up to create what is effectively a virtual circuit within the connectionless network. This allows individual nodes to reserve resources for that 'connection'.
IBM and Technical University of Munich team demonstrate how Shor's algorithm, which can't be cracked by conventional computers, can be solved quickly with quantum computing
Hubble Space Telescope finds superflares from young red dwarfs could strip away planetary atmosphere
Younger stars are 100 to 1,000 times more energetic than when they're older
Two of the big four supermarkets will use the system to control sales of restricted products
PUBG news and updates: November's Update #23 to bring new Skorpion pistol and changes to blue zone visibility
Genuinely useful side-arm coming to PUBG in Update #23