HFT Does Not Make Sense Anymore - by Henrique Bucher

Back in the day, when I worked in high-performance computing, we invested heavily in network cards that had low latency characteristics. HPC applications rely on the Message Passing Interface (MPI) for communication between parallel processes, where multiple nodes work together to solve complex computational problems. This was in the early 90s known as a “Beowulf cluster” or simply “cluster”. Low-latency network cards were well-suited for MPI-based communication since the faster you get the computation results from one node to another node waiting for it; you maximize the amount of time that CPUs are kept busy. Myricom, a company known for its high-performance networking solutions, offered network adapters that were particularly good at this. People get shocked when I point out that a half-round trip latency around the year 2000 with a Myricom card was about 4 microseconds. Low Latency Trading Insights is a reader-supported publication. To receive new posts and support my work, conside

saved by

• Tony Leapo