Re: 2.2.16 and large process failures

[Marcelo Tosatti]

On Tue, 27 Jun 2000 [e-mail] wrote:

> I run a number of large processes, consuming 1 GB of real
> memory and up 4 GB of swap, that run for up to 3 days each.
> 
> A machine upgraded from 2.2.15 to 2.2.16 is no longer able
> to complete the larger of the tasks.
> 
> The particular machine has 896MB of real memory and
> 8GB of swap.
> 
> Memory: 906776k/917440k available
> 
> % swapon -s
> Filename                        Type            Size    Used    Priority
> /dev/hdc1                       partition       2056280 0       1
> /dev/hdc2                       partition       2056312 0       2
> /dev/hdc3                       partition       2056312 0       3
> /dev/hdc4                       partition       2080408 0       4
> 
> The four partitions listed above are 2GB each as shown, interestingly
> top only reports:
> 
> Swap: 4055008K av,      0K used, 4055008K free
> 
> Don't know why the difference.
> 
> In any case, runs under 2.2.16 die of memory exhaustion rather
> before they should.
> 
> A graph of the event is available at:
> 
>     http://www.litzinger.com/memory.png           (3K)
>     http://www.litzinger.com/memory.jpg           (25K)
> 
> (scratch out the % on the charts)
> 
> The application had just released a few hundred MBs and was in the
> mist of allocating lots more.
> 
> It looks like the application was just about to go over 1GB of
> allocated memory, though it had already gone beyond the 896MB of
> real memory, and had been using the swap for the previous 6 hours.
> 
> The system is an Athlon 533 (all in spec).  The kernels are compiled
> with the 1GB maximum physical memory model.
> 
> The machines running 2.2.15 can run to completion using all of memory and
> 1 to 3GBs of swap.

Almost all 2.2.17pre patches had changes in VM.
In my tests, 2.2.17pre8 seems to be pretty stable.

Could you please try to run your big processes in 2.2.17pre8 and report
us results?  

Thanks