Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 10 Nov 2006 13:47:31 -0800
From:      Milo Hyson <milo@cyberlifelabs.com>
To:        freebsd-java@freebsd.org
Subject:   Re: Performance of Java on FBSD vs. others...
Message-ID:  <DC452F20-1AA1-4254-9160-22985B740370@cyberlifelabs.com>
In-Reply-To: <20061110213313.GA72658@ace.b020.ceid.upatras.gr>
References:  <20061110203714.GA89006@ace.b020.ceid.upatras.gr> <20061110124459.M88944@turing> <20061110213313.GA72658@ace.b020.ceid.upatras.gr>

next in thread | previous in thread | raw e-mail | index | archive | help
Low-level benchmarks of the systems (e.g. bytebench) might help shed  
some light on things. I once found the apm module was dragging down  
my context-switching performance by a huge margin. I'm not saying  
that's the issue here, but eliminating some sort of low-level issue  
would help to narrow down the possibilities.

--
Milo Hyson
CyberLife Labs


On Nov 10, 2006, at 13:33, Nikos Ntarmos wrote:

> Hi Nick.
>
> On Fri, Nov 10, 2006 at 12:54:25PM -0800, Nick Johnson wrote:
>> I suspect that even with only two threads, the threading
>> implementation you use may make a difference.  Check your
>> /etc/libmap.conf and if it's easy to do, try again with the alternate
>> thread implementations (libthr, libc_r) just for kicks.  Refer to the
>> man page for libmap.conf for an example.
>
> I'm using libthr, since libpthread has some stability problems for me.
> Moreover, I've seen people report libthr as the fastest of the tree
> implementations, so i thought i'd stick to that. Anyway, I also tested
> against libpthread without any noticeable difference. There's no  
> libc_r
> on my system, so it's not an option.
>
>> I'm not sure if the JVM uses shared memory segments, but you might
>> want to sysctl -w kern.ipc.shm_use_phys=1 to keep shared memory from
>> using swap.
>
> Did that too. No noticeable difference, though...
>
>> How are you setting your heap size?  There could be performance
>> differences between the memory allocators on different operating
>> systems. What happens if you set -Xms and -Xmx to the same value (eg,
>> 1024m) to get all of the allocation to happen up front?
>
> I was using -Xms256m -Xmx1024m. Tried with -Xms1024m -Xmx1024m, again
> without any noticeable difference.
>
>> You might also play with /etc/malloc.conf if you find quirks of  
>> memory
>> allocation.
>
> Malloc options are at their default (i.e. no /etc/malloc.conf).
>
>> Are you using -client or -server?
>
> -client (the default). Also tried with -server, without any noticeable
> difference.
>
>> What about default GC tuning options? These can be platform- 
>> dependent.
>> You may want to explicitly set these to compare apples to apples.
>
> My main consideration is not the fairness of the comparison but the
> difference in speed. I've tried all possible -X combinations but
> (guessed it?) without any noticeable difference; FreeBSD remains the
> lagger...
>
>> I can't think of any other reason there'd be such a difference  
>> between
>> OSes.  Maybe someone else has some ideas.
>
> I was thinking something along the lines of "there are bits and pieces
> of jre/jdk implemented in native code and fbsd doesn't have'em  
> yet", or
> (hopefully) "you have to sysctl-turn a.b.c on"...
>
> I'm really interested in hunting this down, since I'm going to spend a
> considerable part of the next few weeks running even larger  
> versions of
> the program in question, and switching back and forth between freebsd
> (OS-of-choice for coding) and win32 (apparently the fastest of the
> three in executing the code) is a major pain (and loss of time).
>
> Thanks again.
>
> \n\n
> _______________________________________________
> freebsd-java@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-java
> To unsubscribe, send any mail to "freebsd-java- 
> unsubscribe@freebsd.org"
>




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?DC452F20-1AA1-4254-9160-22985B740370>