Re: [Samtools-help] Picard mark duplicates memory usage

[Alec W. <al...@br...>]

Hi Sendu,

I don't know the relationship btw Java heap size and overall memory 
overhead.  I suspect it depends on a lot of factors.

-Alec

Sendu Bala wrote:
> On 27/04/2010 16:08, Alec Wysoker wrote:
>> Hi Sendu,
>>
>> Yes, I was being dense. I don't know what your hard memory limit is
>> relative to what you pass to Java via -Xmx parameter. At Broad we pass
>> -Xmx2G, and have a hard memory limit of 10G. You probably don't need a
>> 5x multiple, but I suggest having the hard memory limit be at least
>> 2-3GB larger than the -Xmx value to account for JVM overhead.
>
> For all of 1000 genomes I've found just 1GB greater was good enough, 
> which is why this one caught me by surprise. The standard overhead 
> should be constant, shouldn't it? And if then uses more than normal, 
> it's because of an inaccurate guess, as you suggested before?
>
>