What do Computer Engineers do?

What are some differences between approaches to low latency network/communication problems in high-frequency trading between electrical engineers and computer scientists?

• I'm assuming that engineers are more focused on hardware, materials properties and signal processing and computer scientists on developing the associated software / algorithms. Or is there much difference in practice and do firms generally require both disciplines to excel in HFT?

• Answer:

  A good electrical engineer typically tends to be fairly good low-level language programmers, and a good programmer usually knows what he needs to know about computer architecture/networks, so I wouldn't say that either one can only approach low-lat from a single perspective. However, there are indeed two perspectives: Hardware: 1. Faster data-transmission: people used to use standard cable. Soon it just wasn't fast enough, so they switched to optical fiber. Later it turns out that Microwave is just better because of its straight transmission path. Now microwave is the common practice of HFT shops, and even non-HF algo-trading shops as well. 2. Closer geographical distance: most big shops have computers located right next to the exchanges (this is called "colocationed cluster", or simply "colo"). 3. Simply better machines: "my computer is faster than your computer!" Trading shops are almost always the first ones to make an order when processor vendors comes up with a new architecture or branch predictor design. Software: 1. Write optimized code: good quant developers know their ways around cache-utilization, pipeline optimization, and sometimes a little bit of crazy bitwise hacks. Each of these give you precious microseconds that can earn you millions (or save you from millions of loss). 2. Parallel programming: make use of multiple machines/processor units. Functional programming comes into play here since parallel-computing is just natural with functional prog. 3. Compile-time optimization: turning all optimization options on in a cutting-edge compiler may sound crazy for a consumer product, but here it is necessary. You just can't afford one extra line of assembly code. I would say, now, in 2014, the barrier of entering HFT has become really high and achieving all of the above is almost a must just for you to be IN the game. Profit margin has become much lower as well, due to competition.