Henning Brauer writes: 

> quite nicely shows a remote concurrency higher than 250 slows the process
> down overall. 
> Of course this highly depends on a hughe number of external factors, but

One such huge factor is memory. With some number of concurrency higher than 
250, he might be swapping, and hence the poor performance result. I don't 
see if this happened on that page. I don't see any memory usage statistics 
either. 

Reply via email to