virtualbox.org

Good morning,

I have experienced guest crashes of different guests on two host systems. The problem seems to be related. On a host system with several guests, one of the guests aborts but can be started again.
I've moved to another host and from Virtualbox 5.x to 6.1.4, which means my old system has become obsolete.

Here is a description of my most recent case.

Host system:
HP DL 380 Gen6
8 physical cores, hyperthreading enabled (I know... but it's an entirely private system)
64G RAM
Host OS: Devuan Ascii 2
Virtualbox 6.1.4 r136177 (Qt5.7.1)
Virtualbox Extension Pack present

Running 7 virtual machines, configured with a total of 7 cores and 18GB RAM

Crashed guest:
Windows 7 (64bit)
8GB RAM
2 CPU, VT-x/AMD-V, Nested Paging, PAE/NX, Hyper-V virtualization
This used to run for about a year on a weaker host (DL380 Gen6, too) with Virtualbox 5.x without any trouble.

Logfile is attached.

I am looking at this line in the manual:
https://www.virtualbox.org/manual/ch03. ... alsettings
3.5.2. Processor Tab

You should not configure virtual machines to use more CPU cores than are available physically. This includes real cores, with no hyperthreads.

With 8 physical cores in the host system, does this mean I cannot configure one guest with more than 8 cores, or I cannot configure guests with a total of more than eight cores altogether?

Any help would be greatly appreciated.
Matthias

mksmr wrote:With 8 physical cores in the host system, does this mean I cannot configure one guest with more than 8 cores, or I cannot configure guests with a total of more than eight cores altogether?

It is recommended to keep one core unassigned for the host to use. So technically, the total number of cores allocated to guests should be one less than the total host physical cores. In your case, 7 cores for the guests. One guest with 8 cores is pushing it, more than 8 would probably be a bad idea.

Also, more cores in a guest slows down the guest, due to overhead in Virtualbox scheduling the guest threads for the cores. Modern Windows is happy with two cores, and Windows 10 seems to need two cores. Throwing more cores is only helpful if you have a program installed that can use more cores (parallel-programmed) and needs to use them. A guest running two cores' worth of data will go faster on two cores than on three or four, according to forum gurus.

The guest does not actually "take" the number of cores you set in Virtualbox. It is allowed to use that number of cores when Virtualbox schedules the guest threads to use them. The host still has all eight cores to use. But if the guests totaling seven cores all go full throttle, then the host still has a core to run itself.

However, it is possible to over-provision the guest cores and still have stable running. I had two 2-core guests and a 1-core guest running on a four-core host, all OS's happy and stable. But not all OS's were going full throttle. Each had its bursts but they were low-usage most of the time. If they all went 105% on the reactor, then I'm sure the host would have choked.

I also run 2-core guests on a 2-core hyperthreaded laptop, and the gust & host stay stable. Just don't fold proteins in the guest... I am aware what might happen if the host and guests all want to use the over-provisioned cores, so it's a risk assessment.

mksmr wrote:8 physical cores, hyperthreading enabled (I know... but it's an entirely private system)

I haven't heard that hyperthreading is bad, only that Virtualbox doesn't use it. I have never turned off hyperthreading in a host PC, and I vaguely remember reading once that turning hyperthreading off makes the CPU run slower.

mksmr wrote:Running 7 virtual machines

More than two modern OS's on a platter drive will swamp the drive and kill the OS. Had that happen before. More OS's can run on an SSD, but even SSDs should be swampable at some point. How many host disks do these guests reside on?

The log just cuts off, with a lot of LsiLogic errors, a Guest Additions report (meaning the OS booted) and audio toggling. I don't know if any of this is normal.

I'd guess, after addressing any issues with CPU core assignments and disk positioning, see if there are common log entries in failing guest's logs.

I basically have
- one workhorse (Debian Linux LTSP server for four users, with normally no more that two online at the same time),
- the said Windows 7 machine for precisely one application I'm not getting to run otherwise with only one user
- one Linux guest running OpenKM DMS, with at most two users at the same time
- one OpenBSD firewall/NAT
- one Nagios monitor
- one Tor exit node

The actual work is done on the Debian LTSP, but from what I see it's under no extraordinary load (as long as not three of the four users are trying to stream videos). The Windows 7 guest is sparsely used, and from what I see, the load on the rest of the machines is neglectable. On the old host, I suspected RAM to be the bottleneck (only 24G altogether). The new host has 64G.

All machines reside on a RAID5 array consisting of 4 SAS HDDs (HP 300G, 10.000 rpm).

When the Windows 7 guest crashed, I was in the process of moving, the machines weren't actually in use, and the Debian LTSP guest is still running on the old host.

Just to make sure, I have reduced the number of configured cores to 6 now. Maybe I'll swap another two machines back to the old host, but in the long run, I don't want to have both running.

Another crash - the guest wouldn't start on host reboot and crashed. But it started without any problems with "VBoxHeadless".

Logfile is attached. I must admit it's pretty much Chinese to me.