Gc performance?

Google Group
1

hi,

anybody have much experience / benchmarking / results regarding the
impact gc has when used? e.g. j. harrop always claimed Boehm was just
Bad News, especially for anything not single-threaded.

thanks.

2

I did compare Boehm-GC with the GC I wrote for ATS1 (not ATS2) on a few
small
examples. For instance, the following one:

Boehm-GC is a generic GC; it is quite often to encounter cases where
Boehm-GC is 100% or
even 200% slower than ATS1-GC. But this is not really an issue for
Boehm-GC in practice.
Actually, I will be really glad if someone could tell me scenarios where
one must rely on a faster GC
than Boehm-GC.

By the way, one really really really good thing about Boehm-GC is its
portability. This is the
primarily reason as to why I moved GC out of ATS2.On Thursday, July 10, 2014 9:00:20 PM UTC-4, Raoul Duke wrote:

hi,

anybody have much experience / benchmarking / results regarding the
impact gc has when used? e.g. j. harrop always claimed Boehm was just
Bad News, especially for anything not single-threaded.

thanks.

3

If I need to deal with hard real-time constraints, then I will stay away
from GC; I will instead use linear types to support safe memory
allocation/deallocation.

thank you, that sounds like sage advice, since apart from things like
the Azul JVM collector, we don’t really have - as far as i know - good
realtime concurrent collectors easily available all over the place.

i should really really really be getting down and learning to use ATS
with iOS and Android and OpenGLES so i can write games that are
performant and yet have drastically fewer changes of silly memory
bugs.

4

Actually, I will be really glad if someone could tell me scenarios where one
must rely on a faster GC
than Boehm-GC.

video games etc. tend to not work well with any gc of any kind,
especially on smaller mobile type devices. i’ve used various systems
that include gc (java, flash, haxe-openfl) of some sort, and even
sometimes claim to be faster than boehm, but still there seems to
inevitably be stuttering issues. i really wish the world could get to
the point where gcs aren’t so hiccuppy.

5

Well, a GC being faster does not imply its being more responsive.
Usually, it means that the GC takes less overall time to reclaim memory.

I have to say that I have strong doubts about so-called “real-time” GC.
That is why I said no GC is good enough if Boehm-GC is not good enough.

If I need to deal with hard real-time constraints, then I will stay away
from GC; I will instead use linear types to support safe memory
allocation/deallocation.On Friday, July 11, 2014 1:35:43 PM UTC-4, Raoul Duke wrote:

Actually, I will be really glad if someone could tell me scenarios where
one
must rely on a faster GC than Boehm-GC.

video games etc. tend to not work well with any gc of any kind,
especially on smaller mobile type devices. i’ve used various systems
that include gc (java, flash, haxe-openfl) of some sort, and even
sometimes claim to be faster than boehm, but still there seems to
inevitably be stuttering issues. i really wish the world could get to
the point where gcs aren’t so hiccuppy.