Hacker News Reader

This has been submitted to HN a few times before (search: lmax.html)

There was a pretty good discussion on this particular thread: https://news.ycombinator.com/item?id=3173993

An oldie but a goldie. Always thought this was a very elegant solution for high throughput.

I would like to see a Call Me Maybe-style test of the automatic failover from one disruptor to the next. At 6 million TPS I don't see how they could stay exactly in sync. Is the output queue allowed to emit duplicates?

Ordering is enforced by the logging mechanism rather than the business logic processing.

The redundant business logic processing should be happening in parallel continuously with the results being verified identical.

This type of system has the added benefit of continuously checking for hardware failure by checking against multiple execution units.

The LMAX Architecture (2011)