On 22 Feb 2015 at 19:43:08, Aaron Hardy at AC 
(ashtarcommunicati...@gmail.com(mailto:ashtarcommunicati...@gmail.com)) wrote:

> Hi Vincent,  
>  
> Thanks for the follow-up.  
>  
> I never managed to track down the exact cause of the futex lock or 
> gettimeofday calls. My best guess is that it has something to do with XWiki 
> running on an Amazon EC2 instance. I didn't see the same kind of issues on a 
> local machine. I had another thread a few months ago where I asked if anyone 
> had successfully run XWiki on AWS, and it didn't seem like there was much 
> experience with that setup.  
>  
> My short term solution was just to double the RAM and throw it all at XWiki. 
> That seems to have mostly dealt with the stability issues (at least, the 
> instance rarely crashes under load), but under the hood I think it's still 
> performing much worse than it should be.  
>  
> Any ideas you have about what might be causing the issue, or any tips for 
> running XWiki on Amazon under reasonably high loads would be much 
> appreciated. 

Thanks for the information and indeed you could be right about the Amazon EC2 
context for the futex problem.

I don’t have any experience with running XWiki on Amazon EC2 so I cannot help 
much, sorry.

Thanks
-Vincent

> Thanks,  
>  
> aaron  
>  
>  
>  
> On Fri, Feb 20, 2015 at 10:05 AM, 
> vinc...@massol.net(mailto:vinc...@massol.net) wrote:
> > Hi Aaron,
> >  
> > Sorry you didn’t get any answer…
> >  
> > Were you able to find out the problem and improve XWiki under load?
> >  
> > Thanks
> > -Vincent
> >  
> >  
> > On 20 Sep 2014 at 06:52:10, Aaron Hardy at AC 
> > (ashtarcommunicati...@gmail.com(mailto:ashtarcommunicati...@gmail.com)(mailto:ashtarcommunicati...@gmail.com))
> >  wrote:
> >  
> > > Trying to track down the source of some persistent performance issues 
> > > under
> > > load with 6.2 RC 1 (haven't had time to upgrade to 6.2 final yet). I'm
> > > maxing out at about 30 or so users before dying on a system with 4GB of 
> > > RAM.
> > >
> > > The java interpreter is getting pegged to 100% of the CPU, and about 70% 
> > > of
> > > the calls in the stack trace are futex lock errors. A good chunk of the
> > > remainder are gettimeofday calls.
> > >
> > > Potentially useful info (sorry for formatting):
> > >
> > > % time seconds usecs/call calls errors syscall
> > > ------ ----------- ----------- --------- --------- ----------------
> > > 72.75 5361.592109 45075 118947 14045 futex
> > > 12.28 905.064191 466 1943354 gettimeofday
> > > 4.83 356.110745 3709487 96 45 restart_syscall
> > > 2.88 212.173358 836 253729 clock_gettime
> > > 2.71 199.867690 8802 22706 recvfrom
> > > 2.10 154.843342 37841 4092 poll
> > > 1.14 84.282371 580 145307 130153 stat
> > >
> > > Any ideas?
> > >
> > > thanks,
> > >
> > > aaron
> >  
>  

_______________________________________________
users mailing list
users@xwiki.org
http://lists.xwiki.org/mailman/listinfo/users

Reply via email to