the relationship between metadata size and heap size

[Walter Forbes Hoehn (wassa)]

Overly generous, I’m certain, but this is what I use currently:

-server
-Xms512m
-Xmx 4096m

In my fairly extensive load/performance testing of the V3 IdP I haven’t noticed any lengthy GC lag spikes with this amount of RAM.  With RAM at a current average cost of @10.00 US $ per GB, I’d rather not run anywhere near the ceiling unless there is a performance-related reason to do so.

-Walter


> On Sep 14, 2015, at 3:47 PM, Cantor, Scott <cantor.2 at osu.edu> wrote:
> 
> On 9/13/15, 8:37 PM, "Cantor, Scott" <cantor.2 at osu.edu> wrote:
> 
>> I was expressing a personal opinion. I don't know whether 512 is actually insufficient, we'll have to run some tests.
> 
> 512 is probably ok for most of the current metadata sizes around, but the InCommon + edugain aggregate is too big for even 768 in my testing, I'm up to 1G before reloading is successful, and I haven't really tested real IdP functionality to see if there's enough room to actually run in.
> 
> Fixing the logging level makes this easier to see of course.
> 
> So I would say 1G is the minimum for many at this point or will be very soon.
> 
> -- Scott
> 
> -- 
> To unsubscribe from this list send an email to users-unsubscribe at shibboleth.net