Latency is an often-neglected quality when discussing performance. Often, the focus is on bandwidth or throughput; however, a big airplane loaded with disks can provide high bandwidth but terrible latency. Financial data is smaller than multimedia data, but often requires very short response times.

This paper illustrates this point with examples from the financial market. It explains measurements and their pitfalls, and surveys some techniques and constraints involved in reducing latency. The primary example is from the foreign exchange market, where the author’s employer lost business to competitors who could respond faster to quote requests. Once this was resolved by adding a cache, the competitors got faster as well, causing financial losses because other companies would take advantage of the bank’s lagging price quotes, getting slightly older rates that were better than the latest market rates. Whether this is a contribution to a healthy economy is not the topic of the paper.

One cannot optimize without measurement. This is explained rather well with an illustration of techniques and pitfalls. Once measured, optimization proceeds to remove bottlenecks, measuring the results without neglecting the global view of system architecture and composition. Topics such as caching, cache warming, locality, protocol selection, storage, and redundant processing are reviewed.

This paper gives background, ideas, and inspiration, but not enough details to implement the solutions. Therefore, I recommend it as a very accessible survey of latency (not only in finance).