[Live-devel] Benchmarking

[Vlad Seryakov]

Ross Finlayson wrote:
>> Not sure these results are representative enough but looks like overhead
>> in BasicTaskScheduler is pretty significant.
> 
> More accurately, the overhead (might be) 'pretty significant' if you 
> happen to be using an OS that implements "select()" inefficiently. 
> I.e., any inefficiency would be a fault of the OS, not an inherent 
> problem with the "BasicTaskScheduler" code.

Sure, i meant exactly that.

Also it is common knowledge that select/pool behave poorly in case if 
large number of sockets, at least in Linux, that's why epoll was 
introduced and kqueue in FreeBSD world. Solaris poll is more effective 
than in Linux. Sadly, epoll does not handle disk files.