Another precision, I have removed all gc tuning values I used before having 
this problem, so the vm has all the default values for gc, I don't know if 
it is gc or memory related.

"Alain_Rastoul" <alr.dev at free.fr> a écrit dans le message de news: 
i9sch8$kuh$1 at dough.gmane.org...
> Thanks Levente, but. . . no, wrong guess , sorry :-)
>
> Instead of memory intensive I should have said  keeping memory about 95/99 
> % used all the time until hang.
>
> When squeak hangs it is about 60, 80, 90 or 100Mo , not always the same, 
> but not that much because it is a small test database (less than 500 mo in 
> mssql) and in most databases a lot of columns have redundant values - ie 
> bad selectivity.
> I've been thinking of something wrong with FFI/ODBC, but another test I 
> just made without it have the same problem (initially with Cog but I 
> reported it in my squeak image) : I load a 1Gb bulk export file of the 
> table of another database  into one big dictionary.
> With squeak It hangs in few minutes at about 60 mo , in Cog it finishes in 
> 30 minutes at about 220 mo.
>
> I'm trying to compile the vm (ok for now) and run it under debug  but gdb 
> is not ok at this time, I have problems with my mingw toolchain (MSVC 
> didn't want to compile perhaps due to the express version I have), I 'm 
> not sure it will be useful and I would prefer a "smalltalk" way to search.
>
> I don't understand why everything seems to be ok without MessageTally ?
> It seems abnormal to me - am I wrong? - to see only few processes in the 
> debug call stack ?
> Another symptom that makes me think processes are terminated is that in 
> the debug output console a lot of event buffer overflow messages are 
> displayed when I move the mouse , I suppose because the UI process has 
> vanished and does not peek them.
> I tryed to intercept process termination but it didn't work, I will make 
> another try with dump to a file it could work if  they are "cleanly" 
> terminated
>
> Thank you BTW for your response, any ideas are very welcome
>
> Alain
>
> "Levente Uzonyi" <leves at elte.hu> a écrit dans le message de news: 
> Pine.LNX.4.64.1010221455550.13297 at login03.caesar.elte.hu...
>> On Thu, 21 Oct 2010, Alain_Rastoul wrote:
>>
>>> Hi Everybody,
>>> I have a standard squeak 4.1 with seaside 3.0 preloaded under Windows XP
>>> (seaside 3.0 rc OneClick Image) with FFI, ODBC packages installed and a 
>>> very
>>> memory intensive application who load data from a MSSQL server and hash 
>>> it
>>> into dictionaries.
>>> It is running in a process started at systemBackgroundPriority, and is
>>> launched with a MessageTally spyAllOn:.
>>> After running a few minutes it hangs up at full CPU, and I can't see 
>>> what is
>>> going on, everything
>>> is hung (no Ctrl., no mouse)
>>> When I open the debug console and dump processes I see only few 
>>> processes
>>> (MessageTally>spyEvery:on:,
>>> handleTimerEvent, userInterruptWatcher, lowSpaceWatcher) but no UI 
>>> process
>>> and no app process.
>>> I tryed to put halts in code but it doesn't change anything.
>>> I noticed that it occurs when the memory usage is about 99% (I see it 
>>> with a
>>> MemoryUsageMorph ticking),
>>> but not allways at the same time, it can load 10k, 20k or 30k rows 
>>> before
>>> hanging.
>>> If I remove the MessageTally, it ran several times without hanging.
>>> I don't see any direct relation between MessageTally and hanging the VM
>>> (allocations / GC ?)
>>> I will try to generate and compile the vm with MSVC if I can (I can't 
>>> test
>>> my app under linux)
>>>
>>> What is the way go with this kind of problem ?
>>> Any help, tip for debugging, advice, idea, remark (...) are welcome.
>>> Thanks in advance
>>
>> I guess you're running out of memory. What's the memory usage of Squeak 
>> when the image hangs?
>>
>>
>> Levente
>>
>>>
>>> Regards,
>>> Alain
>>>
>>>
>>>
>>>
>>>
>>
>>
>
>
>
>
>