 |
BorlandTalk.com Borland discussion newsgroups
|
| View previous topic :: View next topic |
| Author |
Message |
Mike Guest
|
Posted: Mon Nov 06, 2006 9:11 am Post subject: FastMM4 - lockup |
|
|
Hi
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
application. Test it well. We started using it in our last version of
our server. Under heavy load
the server just locked the requests stopped being processed the cpu = 0
usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the memory
manager - it just appeared to be as we were able to replicate the problem.
General info - > Each request would create memory of about 1M
Mike |
|
| Back to top |
|
 |
Jouni Turunen Guest
|
Posted: Mon Nov 06, 2006 9:11 am Post subject: Re: FastMM4 - lockup |
|
|
Hi Mike,
| Quote: |
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
application. Test it well. We started using it in our last version of
our server. Under heavy load
the server just locked the requests stopped being processed the cpu = 0
usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the memory
manager - it just appeared to be as we were able to replicate the problem.
General info - > Each request would create memory of about 1M
|
Can you tell more details about the pc where you tested, mainly
the exact CPU details? Was it something with four cores?
Regards,
Jouni
--
The Fastcode Project: http://www.fastcodeproject.org/ |
|
| Back to top |
|
 |
Mike Guest
|
Posted: Mon Nov 06, 2006 9:53 pm Post subject: Re: FastMM4 - lockup |
|
|
Hi
The CPU is Server with 2 CPU's and hyperthreading turned OFF.
The system is sending requests to the database which causes the large
allocation routines
in fastmm4 to be called.
Mike
"Bart van der Werf" <bluelive (AT) xs4all (DOT) nl> wrote in message
news:454f0abc$1 (AT) newsgroups (DOT) borland.com...
| Quote: |
"Jouni Turunen" <jouni.turunen (AT) no (DOT) spam.xenex.fi> wrote in message
news:454eeb44$1 (AT) newsgroups (DOT) borland.com...
Hi Mike,
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
application. Test it well. We started using it in our last version
of
our server. Under heavy load
the server just locked the requests stopped being processed the cpu = 0
usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the
memory
manager - it just appeared to be as we were able to replicate the
problem.
General info - > Each request would create memory of about 1M
Can you tell more details about the pc where you tested, mainly
the exact CPU details? Was it something with four cores?
Having tested with 4 cores and not finding a deadlock that i could
traceback
to fastmm i have faith in it, however does the fastmm code use a simple
lock
priority scheme to ensure no deadlocks ?
|
|
|
| Back to top |
|
 |
Jouni Turunen Guest
|
Posted: Mon Nov 06, 2006 10:02 pm Post subject: Re: FastMM4 - lockup |
|
|
Hi Mike,
There's some info about 4+ CPU / "NeverSleepOnThreadContention":
http://sourceforge.net/forum/forum.php?forum_id=587429
Probably your problem is somewhere else.
Regards,
Jouni
| Quote: | Hi
The CPU is Server with 2 CPU's and hyperthreading turned OFF.
The system is sending requests to the database which causes the large
allocation routines
in fastmm4 to be called.
Mike
"Bart van der Werf" <bluelive (AT) xs4all (DOT) nl> wrote in message
news:454f0abc$1 (AT) newsgroups (DOT) borland.com...
"Jouni Turunen" <jouni.turunen (AT) no (DOT) spam.xenex.fi> wrote in message
news:454eeb44$1 (AT) newsgroups (DOT) borland.com...
Hi Mike,
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
application. Test it well. We started using it in our last version
of
our server. Under heavy load
the server just locked the requests stopped being processed the cpu = 0
usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the
memory
manager - it just appeared to be as we were able to replicate the
problem.
General info - > Each request would create memory of about 1M
Can you tell more details about the pc where you tested, mainly
the exact CPU details? Was it something with four cores?
Having tested with 4 cores and not finding a deadlock that i could
traceback
to fastmm i have faith in it, however does the fastmm code use a simple
lock
priority scheme to ensure no deadlocks ?
|
--
The Fastcode Project: http://www.fastcodeproject.org/ |
|
| Back to top |
|
 |
Mike Guest
|
Posted: Mon Nov 06, 2006 10:26 pm Post subject: Re: FastMM4 - lockup |
|
|
May be
But the problem does not appear with a different memory manager
Mike
"Mike" <mike (AT) oodesigns (DOT) co.uk> wrote in message
news:454ee76d$1 (AT) newsgroups (DOT) borland.com...
| Quote: | Hi
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
application. Test it well. We started using it in our last version of
our server. Under heavy load
the server just locked the requests stopped being processed the cpu = 0
usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the
memory
manager - it just appeared to be as we were able to replicate the problem.
General info - > Each request would create memory of about 1M
Mike
|
|
|
| Back to top |
|
 |
Pierre le Riche Guest
|
Posted: Tue Nov 07, 2006 1:46 am Post subject: Re: FastMM4 - lockup |
|
|
Hi Mike,
| Quote: | usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the
memory
manager - it just appeared to be as we were able to replicate the problem.
|
FastMM has much better concurrency than many other memory managers. It most
likely exposes a bug in the application that may not be apparent with WinMem
(which is incidentally near the bottom of the heap with regards to speed).
Regards,
Pierre |
|
| Back to top |
|
 |
Simon Guest
|
Posted: Tue Nov 07, 2006 2:45 am Post subject: Re: FastMM4 - lockup |
|
|
Hi,
I got the exact same problem.
At first I ran a busy multithreaded search server using the Delphi 2005
memory manager without problems.
Then I switched to FastMM 4.66, and the search server started locking-up
after, what seems, a random
period of time. Sometimes, it would take minutes, sometimes up to 24 hours
time before the lock-up.
Sometimes under heavy server load, sometimes not.
After replacing FastMM 4.66 with the FastMM based MM in Delphi 2006, all
locking problems disappeared.
I experienced the exact same locking behaviour as Mike.
The search server it still running, but just doesn't reply to any
communication from the outside world, such as a ping, etc..
The system specs is:
* Dual Xeon @3.2MHz
* 4GB memory, where the server uses close to 1.5GB.
* Running Windows 2k3 32bit Server edition.
Unfortunately I haven't been able to figure out how FastMM is related to the
lock-ups, since there doesn't seem to be any
certain way to provoke the lock-up, so this email doesn't help much :/ Just
wanted to let you know that there *seem* to
be some code in FastMM (at least v4.66) that somehow provokes app lock-ups.
Regards
Simon
"Mike" <mike (AT) oodesigns (DOT) co.uk> wrote in message
news:454ee76d$1 (AT) newsgroups (DOT) borland.com...
| Quote: | Hi
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
application. Test it well. We started using it in our last version of
our server. Under heavy load
the server just locked the requests stopped being processed the cpu = 0
usage etc.... Changing
in back to the old one fixed it (winmem). I am not saying it is the
memory
manager - it just appeared to be as we were able to replicate the problem.
General info - > Each request would create memory of about 1M
Mike
|
|
|
| Back to top |
|
 |
Pierre le Riche Guest
|
Posted: Tue Nov 07, 2006 5:45 am Post subject: Re: FastMM4 - lockup |
|
|
Hi Simon,
| Quote: | Sometimes under heavy server load, sometimes not.
After replacing FastMM 4.66 with the FastMM based MM in Delphi 2006, all
locking problems disappeared.
|
Would you please test something for me? Could you please try FastMM 4.48 and
see if it is stable with your application? You can download it from here:
http://sourceforge.net/project/showfiles.php?group_id=130631&package_id=150399&release_id=369863
I made some changes from 4.48 to 4.50 that diverged the code paths of the
BDS MM and FastMM. Perhaps I broke something... (although I did test it
fairly well).
Thanks,
Pierre |
|
| Back to top |
|
 |
Pierre le Riche Guest
|
|
| Back to top |
|
 |
Simon Kissel Guest
|
Posted: Tue Nov 07, 2006 5:39 pm Post subject: Re: FastMM4 - lockup |
|
|
Pierre,
| Quote: | I made some changes from 4.48 to 4.50 that diverged the code paths of the
BDS MM and FastMM. Perhaps I broke something... (although I did test it
fairly well).
|
Just some feedback: We are using FastMM 4.72 on very heavy loaded servers
with 2 CPUs and on some multicore servers. We haven't experienced any
lockups.
Simon |
|
| Back to top |
|
 |
Mike Guest
|
Posted: Tue Nov 07, 2006 10:44 pm Post subject: Re: FastMM4 - lockup |
|
|
Hi
Sorry - I am sticking with winmem - slow but it works
Version with problem = 4.72
Mike
"Pierre le Riche" <pleriche (AT) hotmail (DOT) com> wrote in message
news:454fc9bc (AT) newsgroups (DOT) borland.com...
|
|
| Back to top |
|
 |
Simon Guest
|
Posted: Wed Nov 08, 2006 2:33 am Post subject: Re: FastMM4 - lockup |
|
|
Just to avoid confusion: I'm not Simon Kissel
___
Pierre,
It's a production server used by thousands of paying customers, so I'm not
too keen of beginning
to experiment with it. But as soon as I get the time, I'll develop a test
project which mimics
the search server under heavy load, and see if I can provoke a lock-up with
that, using FMM v4.66
and FMM v4.48.
Cheers
Simon
"Pierre le Riche" <pleriche (AT) hotmail (DOT) com> wrote in message
news:454fc948 (AT) newsgroups (DOT) borland.com...
| Quote: | Hi Simon,
Sometimes under heavy server load, sometimes not.
After replacing FastMM 4.66 with the FastMM based MM in Delphi 2006, all
locking problems disappeared.
Would you please test something for me? Could you please try FastMM 4.48
and see if it is stable with your application? You can download it from
here:
http://sourceforge.net/project/showfiles.php?group_id=130631&package_id=150399&release_id=369863
I made some changes from 4.48 to 4.50 that diverged the code paths of the
BDS MM and FastMM. Perhaps I broke something... (although I did test it
fairly well).
Thanks,
Pierre
|
|
|
| Back to top |
|
 |
Pierre le Riche Guest
|
Posted: Wed Nov 08, 2006 9:12 am Post subject: Re: FastMM4 - lockup |
|
|
Hi Mike,
| Quote: | Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
|
I found a problem in the segmented large block handling code that could
cause the problem you describe. I have fixed it and will release FastMM 4.74
sometime in the course of the next week. In the meantime I have posted a
bugfixed FastMM 4.72 (4.73) to attachments.
Please accept my sincere apologies for assuming the problem lies in your
code. It has become a knee-jerk reaction of mine, since I do get blamed for
a lot of memory management issues and 99%+ of the time it turns out I'm
innocent - but not this time.
Regards,
Pierre |
|
| Back to top |
|
 |
Bart van der Werf Guest
|
Posted: Wed Nov 08, 2006 4:51 pm Post subject: Re: FastMM4 - lockup |
|
|
"Pierre le Riche" <pleriche (AT) hotmail (DOT) com> wrote in message
news:45519e28 (AT) newsgroups (DOT) borland.com...
| Quote: | Hi Mike,
Just to let this group be carefull when using Fastmm4 lastest ver in a
multi - threaded server
I found a problem in the segmented large block handling code that could
cause the problem you describe. I have fixed it and will release FastMM
4.74 sometime in the course of the next week. In the meantime I have
posted a bugfixed FastMM 4.72 (4.73) to attachments.
Please accept my sincere apologies for assuming the problem lies in your
code. It has become a knee-jerk reaction of mine, since I do get blamed
for a lot of memory management issues and 99%+ of the time it turns out
I'm innocent - but not this time.
|
Well atleast you found the bug :)
Any chance of a binary release of 4.73 before next week ?
What is the last known good version of the binary releases ?
grt, Bart |
|
| Back to top |
|
 |
Pierre le Riche Guest
|
Posted: Wed Nov 08, 2006 5:50 pm Post subject: Re: FastMM4 - lockup |
|
|
Hi Bart,
| Quote: | Any chance of a binary release of 4.73 before next week ?
|
I can do a binary release now, but there won't be any new features - just
this bug fix. I was hoping to spend some time catching up on the feature
requests over the week-end. I know you will be annoyed if I make a new
release now and the improvements you sent me aren't in there .
| Quote: | What is the last known good version of the binary releases ?
|
I added the "segmented large blocks" feature in 4.50. That makes 4.48 the
last good version. The bug is not in the BDS memory manager, and
FullDebugMode is also unaffected.
Regards,
Pierre |
|
| Back to top |
|
 |
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
|