Electronic trading, from an investor’s perspective, is a one-click buy and sell process. But the network behind e-trading platforms that matches the buy and sell orders cannot be more complex.
“This is where they get the orders to buy and sell. This is where the decision is being made,” said Ron Nevo, chief technology officer of cPacket, at the recent Networking Field Day event in California.
A Network to Match Buyers and Sellers in Real-Time
Exchanges hold the power to influence the economy. Traders and brokers across the globe make transactions trading securities round the clock in various marketplaces.
A multi-cast IP network is used to send data downstream to the investors.
“Each trader is registering for a specific stream that carries specific movements or tickers that they care about. They will get the information and make the decision to buy or sell based on this information,” Nevo explained. “There are no retransmissions. You either get a packet or you don’t get it.”
Market movements occur every second. Hence, every packet of data traveling through the network must reach its destination, without fail or delay. At the bare minimum, this requires a lossless and ultra-low latency network. But there is a wrinkle.
When a data packet is transmitted from one interface to another, it makes several hops on the way before it arrives at the destination. During these hops, data is temporarily held in the memory before being sent on its way, a phenomenon called buffering.
In an extremely low-latency network, the aim is to send as much data as possible in the shortest time. To achieve that, large number of packets are queued and transmitted in rapid bursts. This causes short spikes in the traffic, often referred to as microbursts.
Microbursts lead to buffer overflows resulting in packet loss. The bursts create frequent bottlenecks making packet delivery slow and intermittent.
The situation is like an infant chugging milk too quickly to keep up with a fast flow. Some of the milk goes down, and some of it flows back out. This effect, in electronic trading scenarios where every packet matters, can spiral into big losses.
With low-latency switches transmitting GBs of data every minute, buffer overrun occurs in milliseconds.
A Neat Trick
Remedying this problem demands a new delivery architecture that prevents bottlenecks by default. cPacket has a uniquely distributed design that puts programmable ASICs and FPGAs together.
This purpose-built hardware is distributed across all ports turning them into FPGA-based smart ports. With dedicated silicon and FPGAs present on every port, packet processing happens at wire speed simultaneously on the ports. Traffic flows directly from the port to the FPGAs, without ever touching the CPU.
cPacket’s Advanced Packet Processing
A combination of cPacket solutions delivers high-resolution packet capture, and line-speed packet delivery: cVu, a packet broker, cStor, a packet capture appliance, and cClear, a visualization dashboard.
Through REST APIs, cVu and cStor report data back to the central command and control software. cClear exposes the analytics about the nature of microbursts on its dashboard. These include the size of the burst, what caused it, frequency and length, and so on.
Not all microbursts lead to packet loss. So cClear does not automatically create an alert every time a microburst is detected. However, it provides administrators the flexibility to configure an alert system, if they prefer it.
Both the cPacket packet broker and packet capture solutions are agentless, and can be run anywhere, no specialized software required.
The products run lossless packet capture and inspection continually in the hybrid multi-cloud. The analytics they bring home give operators a chance to inspect and identify unusual traffic patterns, and carry out speedy root cause analysis in high-frequency trading networks.
For more, be sure to check out cPacket’s presentations from the recent Networking Field Day event at the Tech Field Day website.
