I still can't wrap my head around how the conventional wisdom in the industry to work around that problem is to add even more slow network I/O dependencies.