[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [RFS] jalview



Hi Pierre,

-jvmmemmax=... has already been mentioned, but just to add that you can use both

-jvmmemmax=512M  -jvmmempc=50

to limit the percentage of machine memory to 50%.  What will happen here is the smaller of the two values will be used (so the value given to -jvmmemmax acts like a cap on the percentage value).

I should also add though that 512MB is left for the OS (so -jvmmempc=100 will not use 100%) but perhaps more importantly there's a minimum value of 512MB reserved for the application (unless machine physical memory is less than that in which case it just tries to grab it all!).  Not sure I've explained that very well, but the above 512M and 50% should be okay for tests.

Thanks again,

Ben


From: James Procter (Staff) <J.Procter@dundee.ac.uk>
Sent: 07 February 2021 15:34
To: Pierre Gruet <pgtdebian@free.fr>; Benedict Soares (Staff) <b.soares@dundee.ac.uk>
Cc: Debian Med Project List <debian-med@lists.debian.org>
Subject: Re: [RFS] jalview
 
>> a case of adding -
>> -jvmmemmax=512M

Thanks for your comments, I was also going to use -jvmmemmax in the
tests. Yet, do you think it will be okay if someone with an i386
computer tries to launch Jalview (I have in mind the main Jalview
application, not my tests)?

Jalview will work, but I do take your point that a pre-i486 machine might struggle for interactive use.  However, i386 is an architecture spec for 32-bit x86 platforms - and still pretty widespread 🙂

> Would it be reasonable to apply a default percentage for the maximum
> heap size of, say, 50 pct instead of 90?  

The problem is that the memory check code Jalview uses doesn't look at the shell's limits - only what's available on the machine, so the percentage calculation is not guaranteed to actually yield a size within the limits specified in the shell. It is safer to specify a hard minimum needed for execution. 

You should be able to run these tests in 128M or even 100M of working memory, but the JVM itself needs some heap .. it's a matter of finding what works I expect. We used to have default minimum of 64MB, and maximum of 256MB for all these test operations.

Jim.


The University of Dundee is a registered Scottish Charity, No: SC015096

Reply to: