> Hi,Since the best performance are usually achieved with #threads = #cores and 
> its recommended to run memcached with #worker_threads = #cores.
> I was wondering why does memcached have a specific thread to accept new 
> connections rather than have the worker threads accept new connections?
>
> One possible explanation I found is the Thundering_herd_problem. But I would 
> imagine that CPU cycles would be wasted on accept only on otherwise idle core 
> which I would Imagine is not an issue for most installations, am I wrong?
>
> I was wondering If any one could elaborate on this? Was it discussed in the 
> past? Is accepting new connections a rare enough event that it doesn't make a 
> difference?
>
> Thanks,
> Ilya

It's been that way for a long, long time. It does help with the thundering
heard problem, and tests over the years have shown that it still performs
better that way.

If some day the opposite becomes true, we either switch back or add an
option.

The foruma is 'worker_threads =~ cores', just pretend the accept thread
doesn't count unless your workload never uses persistent connections.
There're also some background threads that're usually idle and won't count
to that total.

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"memcached" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.

Reply via email to