RE: [pthreads-users] NGPT 2.0.1 with 2.4.18 (RES END)

There will be a new memory allocator in ngpt-2.0.2 that significantly
reduces
the memory footprint, hopefully this will improve the virtual memory use for

your application. In the short term, if you know that it's safe you can use 
the DEFAULTTHREADSTACKSIZE shell variable to reduce the vm overhead per
thread
by reducing the thread stack size.

For example:
	test_str02 -l
creates around 24K+ threads with ngpt-2.0.1. This is definitely virtual
memory
bound. 

With ngpt-2.0.2rc3 it creates around 29K+ threads using the same defaults.
If
DEFAULTTHREADSTACKSIZE=16 is used it creates around 89K+ threads.

NGPT-2.0.2 is expected to be out later this week with the new allocator.

Hope that this helps.

Dave Howell 

-----Original Message-----
From: Peter Waechtler [mailto:[email protected]]
Sent: Monday, September 09, 2002 4:19 PM
To: Howell, David P
Cc: 'Marc-Christian Petersen';
[email protected];
[email protected]
Subject: Re: [pthreads-users] [pthreads-devel] NGPT 2.0.1 with 2.4.18
(RES END)


Am Montag den, 9. September 2002, um 17:29, schrieb Howell, David P:

> Marc-Christian Peterson wrote:
>
>> I replaced the libpthread.so* with the ones in the RPM package, loaded
>> ngptinit, shared mem was set up correct.
>>
>> I've noticed that the following programs (maybe more) are not working 
>> at
> all
>> or not working proberly:
>>
>> - mySQL
>>   /usr/sbin/mysqld: relocation error: /usr/sbin/mysqld: symbol pread64,
>>   version GLIBC_2.2 not defined in file libpthread.so.0 with link time
>>   reference
> [DPH] This was added for NGPT 2.0.2 and is in the release candidates.
> Ngpt_2.0.2
> 	is due out this Wednesday, barring unforeseen problems.

I tried to start xmms - but it eat all my virtual mem.
I also tried to use malloc() instead of mmap() in pth_malloc(), but this
breaks different things in a strange way.

What's wrong with the usage of malloc()? With _REENTRANT it should
be threadsafe - at least I thought this until now ... ;-)