Exegy’s feed-handler model takes 2.5 microseconds from network-card arrival to a normalized update in shared memory. Its simple queue reaches average wait equal to processing time at roughly two-thirds utilization; five million messages a second then calls for about nineteen cores. Pin one unusually busy symbol to one core, though, and the average over eighteen idle neighbors is no consolation. Exegy’s alternative illustration leaves thirty-two cores for feed handling on a sixty-four-core box. That headroom comes directly out of what the strategy can run.

Exegy’s queue model