I hate not understanding things. I joined Metamako after three years in the exchange space, approximately 14 years in private equity before that and 3 years of foreign exchange, commodity and interest rate trading at the very beginning of my career. You’d hope those three years as a trader would position me well to understand how today’s trading gets done. Nope, not even a tenth of it has proven to be relevant. So after joining Metamako as their Chief Commercialisation Officer (I look after finance, marketing and sales), I’ve had to learn about low latency electronic trading and layer 1 switches in particular (in addition to a new vocabulary of acronyms). I figure I am not the only person who could benefit from this basic knowledge.
The OSI Model
The Open Systems Interconnection (OSI) model is a conceptual model that characterizes and standardizes the internal functions of a communication system by partitioning it into abstract layers of functionality.
The lowest layer of the internal functions of a communication system is known as layer 1, the physical layer.
The physical layer consists of the basic networking hardware technologies which transmit data, moving it across the network interface. All of the other layers of a network perform useful functions to create and /or interpret messages sent, but they must all be transmitted down through a layer 1 device, where they are physically sent out over the network.
The main functions of the physical layer are:
Encoding and Signaling: The physical layer transforms the data from bits that reside within a computer or other device into signals that can be sent over the network as voltage, light pulses or radio waves that represent ones and zeroes;
Data Transmission and Reception: After encoding the data appropriately, the physical layer actually transmits the data, and of course, receives it. Note that this applies equally to wired and wireless networks, even if there is no tangible cable in a wireless network.
To summarise then, physical layer technologies are ones that are at the very lowest level and deal with communicating actual ones and zeroes (i.e. sending and receiving) over the network. These devices (prior to Metamako’s MetaConnect) have absolutely no knowledge of the contents of a message. They just take input bits and send them as output. Typically, devices like switches and routers operate at higher layers and look at and act upon the contents of the data packets that they receive.
Still a little confused? An analogy may help!
Computer networks are organised a bit like a traditional business with executives at the top of the stack, managers in the middle and workers at the bottom (the factory floor). Work gets done at the bottom but often it needs to percolate up the layers of management for a decision to be made. This process takes time and is not very efficient but is necessary if the decisions can't be made by the workers.
If we think of the network stack then the bottom layer (layer 1 or the physical layer) of the network does the work of sending and receiving bits across the physical network wires without doing a lot of thinking.
In a conventional network switch, each data message is received at layer 1, then passed up the management chain of the network stack (to layers 2 or 3) where a decision is made. Then, the results of the decision percolate back down the chain to the physical layer to transmit. This all takes time.
The idea of a physical layer switch is that the switching decisions are made in advance so that the work never gets held up. This is called circuit switching and as a result it is approximately one hundred times faster than the conventional switch because the decisions are never required to go above the physical layer.
The physical layer switch isn't as smart or as flexible as a layer 2 or 3 switch but if latency is important then a physical layer switch should give you a significant speed edge.
What is a crosspoint Switch?
O.K., so you’ve gotten your head around what a layer 1 device does but not sure what a crosspoint switch is? I feel your pain or rather confusion.
Crosspoint switches originated in telecommunications around 1915 as a way to create a circuit between two phone lines. Originally these were electromechanical devices that used a metal bar to create an interconnect and therefore were known as crossbar switches. Eventually these were replaced by semiconductor technology but the concept remained the same of having a matrix formed from the set of inputs and outputs.
Modern crosspoint switches are programmable Integrated Circuits (ICs) with the capability to tweak where the signal being communicated goes to. As programmable switches they have become widely used in networks and telecommunications.
You’d think that all there is to trading these days is being fast. Well back in the day, there wasn’t a trader around who knew what the word latency meant. Seriously - least not in Australia. Anyway ...
Latency in a network is the measure of delay from the start of packet transmission to the start of packet reception. The latency of a switch is the measure of how long it takes a packet on the wire or fibre to enter and transit the switch onto the output wire.
If network latency is a major concern then the lowest latency switching solution is a physical layer switch with a latency now measured in single digit nanoseconds compared to hundreds of nanoseconds for many layer 2 switches. This is possible since the latency of the crosspoint switch at the heart of these switches are now specified in picoseconds (parts of a nanosecond).
Crosspoint switches do not allow for the connection of multiple inputs to a single output, that is, they do not allow packets from multiple links to be aggregated into a single link. This means that there is no need to buffer any packets in a queue so they pass through at close to the speed of light.
Another benefit of crosspoints is that they can connect a single input to more than one output at a time without introducing any overhead. This means that a physical layer crosspoint switch can broadcast data to many ports at the same time.
Circuit switching is not for everyone but it turns out that there are range of applications where the low latency is a winner:
Top of rack patch panel in the data centre or test lab
Monitoring of links and bypassing of faults
Data broadcast (i.e. market data broadcast in finance)
Media conversion (e.g convert from optical fibre to copper)
Clock and Data Recovery
Physical layer switches act like a piece of wire but in fact they do a little more than that. They typically have circuitry that regenerates the signal that is passing through. The reason to regenerate is to remove noise from the incoming signal and start again with a fresh clean signal on the output. That is, the individual output bits are replicas of the original input bits where the bits were first generated.
This process is called clock and data recovery or CDR or retiming and is a standard part of all network switches.
There is such a big market for CDR circuitry that many manufacturers produce ICs that perform the function with many other bells and whistles.
MetaConnect includes clock and data recovery - that is, it regenerates a new clean signal. The benefit to users is that it reduces errors in the data and allows the data to be broadcast over longer distances at lower latency.
Given I’m responsible for sales at Metamako, it would be remiss of me not to plug our product whilst I have your attention. What do you mean I basically already have? Just bear with me for another 60 seconds. It will be worth your while ... I promise.
MetaConnect literally takes switch technology to the next level.
MetaConnect is unique in combining ultra-low-latency performance of less than four nanoseconds with high levels of monitoring functionality normally only found on much slower devices. It also provides management statistics and network monitoring with no impact on network latency, improves signal quality and saves rack space.
MetaConnect allows users to replace a significant number of boxes in their server rack: patch panels, taps, signal regenerators, media converters and switches with a single intelligent box which has negligible latency. Each of these devices introduces latency overheads, possible points of failure and ties up valuable rack space (not to mention cost).
MetaConnect has been designed from the ground up using the latest hardware and software components and incorporating a unique combination of features not currently available on the market. The real world benefits are:
< 4 nanosecond latency with virtually undetectable jitter;
broadcasts data in less than 4 nanoseconds – or combining two layers of Connect 48’s, broadcasts to over 2,300 in less than 16 nanoseconds;
a remotely controllable patch panel eliminating the need for physical access or remote hands;
precise timestamping of packets;
connection monitoring and packet statistics to improve network manageability;
converts between different network media such as from fibre to copper;
open standards x86 management platform for future extensibility and customization; and
improves signal integrity on problematic long haul links (MetaConnect regenerates a new signal from the imperfect signal received).
Where network engineers would have been hesitant to introduce an additional network hop for manageability, compliance, monitoring or signal regeneration, they can now do so with virtually no impact on latency. Metamako listened to the market and engineered new levels of performance into the MetaConnect. To characterise it in terms of the OSI model, its more like a layer 1+ device than a layer 1!