Solaris host crash while running VirtualBox

[SommerMR]

Hi VirtualBox-Community,

we use the VirtualBox 4.1.16 on a Sun Fire X4600M2 system (8 AMD Opteron 8356 CPUs, 256GB Memory) running Solaris 10 (uname -a --> SunOS servername 5.10 Generic_147441-27 i86pc i386 i86pc).

For several months this worked well. Troubles started, when we moved from VirtualBox version 4.1.8 to 4.1.16 and applied the recommended Solaris patch set in May 2012.
At that time the troubles could be fixed with patch 147441-19.

With this the system worked nice until I installed the recommended Solaris patch set in December 2012.
Now the host system crashes again - even with a small number of Virtual Machines.
The only change between the working and crashing system is the 'recommended patch set'.

As we only have a support contract for Solaris and not for VirtualBox we don't get support from Oracle.
Of course it is not a problem of the operating system when an application is capable of crashing the whole system.
And if this happens after installing the patch set from Oracle we have to blame the Application (VirtualBox) ... right?
Some years ago, I knew this behavior from MS - but times have changed.
We also use VirtualBox 4.1.16 on a Sun Fire X4600M2 (8 AMD Opteron 8356, 128GB ECC) running CentOS 5.x and do not have these problems.

Back to the issue: anybody else experiencing crashes with Solaris and VirtualBox?
Or anybody knowing about some sort of solution?
Our long term solution will be another operating system. But for the meanwhile any support is highly appreciated.

More details about the crashes are available in the public bug tracker under Ticket #10650

Best,
Jens

[martyscholes]

I had a similar experience last year with different errors:

viewtopic.php?f=11&t=49861&sid=c41b2c0b ... b0ee198a29

Our issues seemed related to networking at the guest level somehow. We know that we can consistently cause a host crash by watching a youtube video on a Windows 8 guest. If / when you figure this out, please let the community know.

By the way, I notice you too are running Sun Rays. If you ever get 3D working on Sun Ray / VirtualBox, please let us know that as well.

Good luck.

[SommerMR]

Hi Marty,

at least the network load due to YouTube, Ebay, etc. is low. Users are not enabled to access most of the 'fun' pages - neither on host nor on guest side.
Indeed the guests (mainly Windows) are restricted to the host file system.
When I checked the community pages, I saw some hints pointing to high network or system load, but I do not think it is really load associated because we experienced the crashes even with as little as four to five users active and the system is (was) running stable with more than 50 users logged in and maybe 10-20 VMs running.

SunRay and ZFS are the last pros for Solaris in our department - most of the rest has been wiped away with the 'O'-company.
3D has never been of relevance for our tasks in the virtual machines, so we never tried.

Best,
Jens

[martyscholes]

You issue looked different than mine. Yours appears to be a pointer error in some allocation code inside vboxdrv. Mine would throw a BUNCH of messages to the log then the machine would quickly slow to a crawl. While we never did pinpoint the issue, it appeared to be network related but not related to network load. We could crash the host with a low network load and certain Windows applications. I recall one guest could crash the host by launching HP's printer management software.

I recall you were running an older build of VirtualBox. Have you considered upgrading VB, even in a zone and testing that?

[SommerMR]

Slow down of the virtual machine is possible, but I do not know, because the VM users normally do not realize this and so they do not report this to me.
The host itself is crashing suddenly. As the internet connection is slow anyway (due to a proxy outside of our reach), we are used to delays and maybe we would not even recognize if it is due to a slow proxy or due to one of the VMs.

If it wouldn't be ECC memory, I would search the problem in the hardware, but we don't get error messages of this kind.
And as it happens after installing the patch set, I don't believe it's only related to the VirtualBox.
I guess the VirtualBox only triggers the crash and the bug is somewhere in Solaris.

I already downloaded the new version of the VirtualBox, but there is no comment on something like our problem included in the release notes, and I don't have the guts to create an additional problem with this.
Right now, I hope that removing the patchset will resolve the problem.

Did you test your installation in a zone? I did not, but if you did so, did VirtualBox only crash the zone or the whole system?
As I do not have a spare X4600M2 and the Sun Try and Buy promotion has stopped long times ago..., it is hard to test without bothering the users.

Best, Jens

[martyscholes]

I never had attempted testing a different version of VB in a zone. The more I think about it, it probably would not work because VB installs kernel modules and the kernel is shared across zones. In fact, I recall issues with VB in zones when the VB versions in zones did not match the global zone.

Pretty much, ignore what I wrote.

[SommerMR]

Thank you for this update.

Anyhow I used the idea to look once more for the option to use a non-antique Linux-Kernel in a Linux branded zone, but it looks as the Linux branded zones stopped with REL 4.
There are some hints on Kernel 2.6.x for OpenSolaris 11 but not for Solaris 10.
At least with VirtualBox on our second X4600M2 running CentOS5 we did not experience any stability problems.

And I hoped that this would hold true for a Linux branded zone, too. What do you think?

Best, Jens