In our prod environment, using quick fix 1.5.2. Recently our upstream set up an algo trading system, it sends lots of DMA orders to our system. When concurrent pressures become high, we found high latency in incoming step and outgoing step.

We add some logs in fromApp() method to record callback timestamp, found between this timestamp and the timestamp of tag35=D have an average of 400ms delay, max delay is 2500ms. At outgoing step, toApp() method have same issue.

In our system, connection 9 sessions, two of them are doing algo trading by DMA, others do manual order trading.