Re: GC performance - GC fragility
- From: Sean Cross <sean.cross@xxxxxxxxxxxxxxxx>
- Date: Thu, 24 Jan 2008 09:03:56 +1300
I am going to put up a followup post on my blog and then retire from the subject as it is taking up far too much of my time.
One of the problems I have had is that people have read more into my posts than I was trying to say (this may be my fault).
These are the things I am not trying to do/say:
I am not trying to say that the Boehm gc will make your delphi apps faster, leak or free.
I am not trying to do a Boehm GC is faster than FastMM competition.
I am not trying to suggest that any gc app is always, or even mostly faster, than than non gc
I am not suggesting that you should add the gc to your existing apps, or even that you should add it to your new apps.
I am not trying to suggest that a gc is perfect, or a better solution in all applications. I can think of several apps, from personal experience , where a gc is a very bad idea.
This is what I was trying to say in my first post:
o A gc has some advantages (expressiveness is the one I am interested in, some lifetime management issues are another)
o A gc has some disadvantages (mem use and performance are 2 that I mentioned)
o Here is a gc for delphi that I have found, have a look if you are interested.
In my second post, my wording should have clearer. I was trying to show that a gc app doesn't have to be slow simply because it as a gc. Given the unquestioning "gc = slow" attitude of many people, I felt a test was needed. I wasn't trying to have a performance bake off which is what it devolved into.
The actual speed of the memory manager is irrelevant to me, and I suspect to most people. I am interested in the speed of the application itself, of which the mm is a part. In my testing, in my usage, the speed differential was not big enough for it to be a major decision factor. But caveats do apply and though testing would be needed before leaping headlong into using gc in a language that doesn't have it.
In my dream world, delphi has a high performance gc (the boehm one is not high performance, it has other design goals such as portability) and people use it or not as the job requires. Maybe Pierre could do it. I won't ask him though.
In the real world, all I wanted to do is raise the profile of gcs in delphi and see if I can get some interest.
My final word of the subject, and then I will actually do the things I am supposed to be doing.
To dismiss garbage collectors as "hideously slow" and thus useless is not helpful or informative. If correct (and it is correct in SOME situations but not ALL situations) then it disqualifies using a gc in SOME roles but not ALL roles.
What is helpful is discussion on when a gc is or isn't appropriate.
Sean
---------------------------------------
Sean Cross
mailto:sean@xxxxxxxxxxxxxxxxxxxx
Pics Print - The photo printing solution for Windows. http://www.picsprint.com
Rental Property Manager - Rental management made easy http://www.sourceitsoftware.com
Blog: http://sourceitsoftware.blogspot.com/
.
- Follow-Ups:
- Re: GC performance - GC fragility
- From: Dennis Landi
- Re: GC performance - GC fragility
- From: Jolyon Smith
- Re: GC performance - GC fragility
- References:
- GC performance - GC fragility
- From: Eric Grange
- GC performance - GC fragility
- Prev by Date: Re: Delphi in a VM ..... VirtualPC 2007, VMWare 6 or ...?
- Next by Date: Re: Garbage collection - performance test
- Previous by thread: Re: GC performance - GC fragility
- Next by thread: Re: GC performance - GC fragility
- Index(es):
Relevant Pages
|
Loading