[bind10-dev] numeric/integer size on CC channel

[Francisco Obispo]

Yes,

I'm all for that, I didn't suggest it was simulated by the BIND10 team, I just meant simulated because it's not a native 64-bit counter. In fact, I agree that it should be handled by something that's already available.

And I agree with Francis, we should make sure that we use an atomic increment. 

Unless you use some sort of shared memory system that stores its own counters on a per-thread basis, and then have another thread do a collection and aggregation of the data for reporting purposes. may sound like a crazy idea but may prevent bottlenecks in case of a large number of threads on busy servers.




On Jun 21, 2013, at 2:32 AM, Michal 'vorner' Vaner <michal.vaner at nic.cz> wrote:

> But even in case we find out there's a case when it could overflow, I think the
> right solution is then to use a library for long numbers and not try to hack
> around the fact the number can get lower at various places scattered around the
> code.