the performance of IO manager in GHC 7.10.1rc2

Bardur Arantsson spam at scientician.net
Fri Jan 30 02:45:46 UTC 2015


On 01/30/2015 02:08 AM, Kazu Yamamoto (山本和彦) wrote:
> Hi,
> 
> As usual, I took benchmark of IO manager to check if preformance
> regression is introduced. Fortunately, I don't see any preformance
> regression.
> 
> +RTS -N<x>    1        2        4        8        16
> ---------------------------------------------------------
> GHC7.8.4       81,413  153,478  270,178  406,448  50,3203
> GHC7.10.1rc2   88,247  148,471  265,392  409,514  49,3890
> 
> Environment: two real 20 cores machines directly connected by 20G
>   - Xeon E5-2650Lv2 (1.70GHz/10cores/25MB) x 2, 8GM, no HT
>   - CentOS 7.0
>   - Two 10Gs are aggregated
> 
> Server: witty -a -m -r 8080 +RTS -A32m -N<x>
> 
> Benchmark: weighttp -n 100000 -c 1000 -k -t 16 http://<ip-addr>/

Out of curiosity: What are the units for numbers in the table? Requests
per second per core?

Regards,



More information about the ghc-devs mailing list