I wonder how much of this comes from the changed semantics of caching in ZFS vs UFS.
UFS used to report space used by the buffer cache as "free" and would release the memory on demand. ZFS reports it as in use but releases on demand. An app written for the former behaviour will see less "free" memory when run on ZFS. Off hand, I don't see a reason that the reporting semantics needed to change, but I think it's too late now. -- Boyd Adamson On Thursday, 10 May 2012 at 9:45 AM, chris.unix.d...@gmail.com wrote: > Either virtualbox needs to be more relaxed about lack of free memory, or zfs > should stop needlessly being so greedy. Either way it's annoying that Oracle > software doesn't play nice with Oracle OS. > I believe an OS is there to service applications, which in this case it's not. > In the meantime, I'll change /etc/system. > > Chris. > Sent via BlackBerry® from Telstra > From: Leigh Maddock <lmaddoc...@gmail.com (mailto:lmaddoc...@gmail.com)> > Date: Thu, 10 May 2012 09:41:06 +1000 > To: <chris.unix.d...@gmail.com (mailto:chris.unix.d...@gmail.com)> > Cc: Melbourne OpenSolaris Users Group<ug-msosug@opensolaris.org > (mailto:ug-msosug@opensolaris.org)> > Subject: Re: [ug-msosug] Brains Trust: Why is VBox not freeing memory after > starting up and stopping a VM? > > I've found it does give priority, it is just some applications try to be > smart and only look for free memory, i believe Oracle dbms is a big culprit > of this, and get the same issues on red hat with disk cache. > Cheers, > Leigh Maddock > Unix Systems Administrator > On May 10, 2012 9:36 AM, <chris.unix.d...@gmail.com > (mailto:chris.unix.d...@gmail.com)> wrote: > > Yeah, I had nobbled zfs in the past, but those changes were lost when I > > could upgrade from opensolaris to solaris 11 and had to do a fresh install. > > > > I have to admit, I'm surprised that higher priority is given to fs caching > > over the priority of an application working correctly. I wasn't expecting > > zfs to stop a vm being started when clearly a few seconds earlier it had > > run fine. > > > > > > Cheers, > > Chris > > Sent via BlackBerry® from Telstra > > From: Leigh Maddock <lmaddoc...@gmail.com (mailto:lmaddoc...@gmail.com)> > > Date: Thu, 10 May 2012 09:31:01 +1000 > > To: Chris Wells<chris.unix.d...@gmail.com > > (mailto:chris.unix.d...@gmail.com)> > > Subject: Re: [ug-msosug] Brains Trust: Why is VBox not freeing memory after > > starting up and stopping a VM? > > > > Your zfs cache is using 4.66gb, if you run top now that should show about > > 3.5gb free depending on what else is running. > > If you want to stop zfs using so much cache to see if it is causing the > > error: > > Add the following to /etc/system > > set zfs:zfs_arc_max = <bytes> > > The reboot. Im sure you can do it online without outage but cant find the > > command, someone else may know. Otherwise might be the easiest way to clear > > the arc cache and test again :). > > 1gb in bytes is probably a good test. > > Cheers, > > Leigh > > On May 10, 2012 12:58 AM, "Chris Wells" <chris.unix.d...@gmail.com > > (mailto:chris.unix.d...@gmail.com)> wrote: > > > crispi@marvin:/VMs/IBM$ kstat -p zfs:0:arcstats:\size > > > zfs:0:arcstats:size 5007987624 (tel:5007987624) > > > > > > > > > On Thu, May 10, 2012 at 12:30 AM, Leigh Maddock <lmaddoc...@gmail.com > > > (mailto:lmaddoc...@gmail.com)> wrote: > > > > Hi Chris, > > > > > > > > Are you running your VDI's off zfs? It's possible the vbox disk is > > > > being loaded into the zfs arc cache. > > > > > > > > Run the following command before and after and see if it is showing > > > > your missing memory: > > > > # kstat -p zfs:0:arcstats:\size > > > > > > > > You can limit the amount of cache ZFS is allowed to use if this is the > > > > case. > > > > > > > > I have seen numerous times where applications try to be too smart for > > > > their own good by checking for free memory as opposed to just asking > > > > for it and letting the OS release the cache. > > > > > > > > Cheers, > > > > Leigh Maddock > > > > Unix Systems Administrator > > > > > > > > On Wed, May 9, 2012 at 10:56 PM, Chris Wells <chris.unix.d...@gmail.com > > > > (mailto:chris.unix.d...@gmail.com)> wrote: > > > > > Thanks Richard. I wouldn't mind, but the behaviour is annoying: > > > > > > > > > > 1) A restart of the same VM will sometimes fail > > > > > 2) The VM will fail half-way through boot up (so I guess the memory > > > > > isn't preallocated). > > > > > > > > > > This kind of indeterminate behaviour is annoying for what should be > > > > > mature products by now. > > > > > > > > > > Chris > > > > > > > > > > > > > > > > > > > > On Wed, May 9, 2012 at 10:35 PM, Richard Smith > > > > > <richard.r.sm...@oracle.com (mailto:richard.r.sm...@oracle.com)> > > > > > wrote: > > > > > > My guess is that the memory is lurking inside kmem caches. Since > > > > > > the system > > > > > > isn't under memory pressure, there's nothing to trigger a reap of > > > > > > them. > > > > > > You can get a better idea where the memory has gone from the mdb > > > > > > dcmds > > > > > > below. There are a lot of kmem caches though, and I expect only a > > > > > > few > > > > > > of them account for most of the "missing" memory. > > > > > > > > > > > > ::memstat > > > > > > ::kmastat -m > > > > > > ::kmem_cache > > > > > > ::kmem_slabs > > > > > > > > > > > > > > > > > > -- > > > > > > ============================================================================ > > > > > > ,-_|\ Richard Smith PAE > > > > > > / \ Oracle Phone : +61 3 8616 > > > > > > 3300 (tel:%2B61%203%208616%203300) > > > > > > richard.r.sm...@oracle.com (mailto:richard.r.sm...@oracle.com) > > > > > > Direct : +61 404 815 885 (tel:%2B61%20404%20815%20885) > > > > > > \_,-._/ 417 St Kilda Road Fax : +61 3 8616 > > > > > > 4500 (tel:%2B61%203%208616%204500) > > > > > > v Melbourne Vic 3004 Australia > > > > > > =========================================================================== > > > > > > > > > > > > > > > > > > > > -- > > > > > Regards, > > > > > > > > > > Chris > > > > > > > > > > _______________________________________________ > > > > > ug-msosug mailing list > > > > > ug-msosug@opensolaris.org (mailto:ug-msosug@opensolaris.org) > > > > > http://mail.opensolaris.org/mailman/listinfo/ug-msosug > > > > > > > > > > > > > > > > > > > > > -- > > > Regards, > > > > > > Chris > _______________________________________________ > ug-msosug mailing list > ug-msosug@opensolaris.org (mailto:ug-msosug@opensolaris.org) > http://mail.opensolaris.org/mailman/listinfo/ug-msosug > >
_______________________________________________ ug-msosug mailing list ug-msosug@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/ug-msosug
