[Mono-dev] gc

Andrea Francesco Iuorio andreafrancesco.iuorio at gmail.com
Mon Oct 20 19:30:06 UTC 2014


If I remember correctly (so wait for someone that can be more precise), except for some debug support, the only real difference between the two implementation is the worker implementation. Nothing that can really justify this difference, if we can assume that you use the same collector and the same version. A so big difference only in the major collections seems to be the difference between the parallel and the serial collector. Maybe in your windows machine you utilise the parallel collector as default ? Can you post more —stats details (the majors data in the GC part can help us to see where is the problem) ?

> On 17 Oct 2014, at 17:30, Neale Ferguson <NealeFerguson at verizon.net> wrote:
> 
> Hi,
> We were helping someone who was having performance problems with a large
> streaming application that they run under .NET and mono. On .NET the app
> takes around 7-10 seconds to do its stuff. On Mono on Linux it was taking
> 47 seconds or so to do the same thing. The mono system had 256G RAM and 48
> cores. When they used Mono on Windows the figure was back to around 7
> seconds. We suggested they use MONO_GC_DEBUG to try and determine where
> they were spending time. It turns out that the old generation scan was
> responsible for nearly all of the difference between the runs (i.e. around
> 40 seconds). How (or does) does the sgen gc differ under Windows to Linux?
> 
> Neale
> 
> 
> _______________________________________________
> Mono-devel-list mailing list
> Mono-devel-list at lists.ximian.com
> http://lists.ximian.com/mailman/listinfo/mono-devel-list



More information about the Mono-devel-list mailing list