Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 17 Sep 2010 14:07:33 +0300
From:      Kaya Saman <kayasaman@gmail.com>
To:        Jeremy Chadwick <freebsd@jdc.parodius.com>
Cc:        freebsd-ports@freebsd.org
Subject:   Re: Fwd: Tomcat6 port keeps locking up??
Message-ID:  <4C934BF5.7@gmail.com>
In-Reply-To: <20100917100723.GA49737@icarus.home.lan>
References:  <4C926418.2050407@gmail.com> <4C9328B9.4010100@gmail.com>	<20100917085621.GA48570@icarus.home.lan>	<4C933284.6050601@icyb.net.ua>	<20100917094212.GA49319@icarus.home.lan>	<4C933A85.8080703@icyb.net.ua> <20100917100723.GA49737@icarus.home.lan>

next in thread | previous in thread | raw e-mail | index | archive | help
On 17/09/2010 13:07, Jeremy Chadwick wrote:
> On Fri, Sep 17, 2010 at 12:53:09PM +0300, Andriy Gapon wrote:
>    
>> on 17/09/2010 12:42 Jeremy Chadwick said the following:
>>      
>>> On Fri, Sep 17, 2010 at 12:19:00PM +0300, Andriy Gapon wrote:
>>>        
>>>> on 17/09/2010 11:56 Jeremy Chadwick said the following:
>>>>          
>>>>> I don't think you understand how Solaris's VM behaves with ZFS.  It
>>>>> behaves very differently than FreeBSD.  On Solaris/OpenSolaris with ZFS,
>>>>> you'll see the ARC taking up as much memory as possible -- but unlike
>>>>> FreeBSD (AFAIK), when a userland or kernel application requires more
>>>>>            
>>>> ^^^^^^^^^^^^^^^^^
>>>>          
>>>>> memory, the Solaris kernel dynamically releases portions of the ARC.
>>>>>            
>>>> Can you please explain that "unlike" part?
>>>>          
>>> When ZFS was first introduced to FreeBSD, I was given the impression
>>> from continual posts on the mailing lists that memory which was
>>> allocated to the ARC was never released in the situation that a userland
>>> program wanted memory.
>>>
>>> An example scenario.  These numbers are in no way accurate given many
>>> other things (network mbufs, UFS and VFS cache, etc.):
>>>
>>> - amd64 system has 2GB physical RAM (assume ~1920MB usable)
>>> - vm.kmem_size="1536M" + vfs.zfs.arc_max="1400M"
>>> - Heavy ZFS I/O results in ARC maxing out at ~1400MB
>>> - Userland application runs, requests malloc() of 1024MB
>>> - Userland gets 384MB from physical RAM, remaining 640MB from swap
>>> - ARC remains at 1400MB
>>>
>>> Is this no longer the case?
>>>
>>>        
>> I am not sure if this has even been the case :-)
>> It is definitely not the case now.
>>      
> I trust your experience with it *much* more than mine.  :-)  It's very
> likely that I'm basing the "ARC remains at 1400MB" claim entirely off of
> what top(1) was showing under either "Inact" or "Wired".
>
> The terminology in top(1) for memory on BSD has always confused the hell
> out of me.  That might sound crazy coming from someone that's been using
> *IX since 1990 and BSD since 1996, but it's true.  The man page does go
> over what's what, but the descriptions are short one-liners (ex.  "wired
> down" doesn't mean anything to me).  This just circles back to my lack
> of knowledge about the VM.
>
>    
Aren't there supposed to be 2 versions of 'top'??

Unix top and Linux top??

Both with slightly different handles on the representation of information?

I just recall reading somewhere!!



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4C934BF5.7>