3 ms·
To be fair, those credit card processors are not running as fast as they could in theory either. You can find another 3 orders of magnitude in certain financia
by bob1029 5y ago
To be fair, those credit card processors are not running as fast as they could in theory either.
You can find another 3 orders of magnitude in certain financial exchanges. In some cases, those millions of transactions per second are being managed on a single thread.
- AccountToUse 5y agoWell, certainly not on Robin Hood's servers.
- nacs 5y agoRobin Hood isn't a credit card processor -- they're stock traders.
- zenron 5y agoLMAX Disruptor is living proof of that. https://martinfowler.com/articles/lmax.html https://martinfowler.com/articles/lmax.html
- bob1029 5y agoThis is precisely what I was referring to. I sometimes wonder if the dev community has blinders on regarding how much elegance lives with this technique.
- anthonygd 5y agoWhy do you suspect blinders? Do you know of many cases where the disruptor pattern should be used by isn't?
- bob1029 5y ago> Do you know of many cases where the disruptor pattern should be used by isn't? The disruptor pattern is the way to solve any problem where you need to process events/transactions/commands/et. al. in a serialized fashion [0] as quickly as possible. This problem space involves almost all business software, databases, simulations, etc. I do not know of any other technique that would allow for faster serial processing of data, especially when controlling for practical applicability. When I refer to blinders, I think I am mostly pointing to the ignorance regarding what is actually possible with the hardware (if you get the software out of its way). Any other technique involving cache-friendly structures, ring buffers, etc. is in the same league as far as I am concerned. [0] To be as clear as possible, "serialized" in this context means that the consequences of event #1 may alter how event #2 would be processed, and so on.
- SilasX 5y agoOh yeah, the exchanges are great on throughput. The two-(business!-)day lag for finality really hurts their latency figures though.