URGENT: Timesync on build 134 with 3.2.6

[hvr_za]

If I rdp to my guest and view the seconds ticking by sometimes 3 seconds would pass before the it updates and then a second later it will jump 3 seconds in 1.5 -2 seconds. My guests are loosing time dramatically. I tried many of the options in the fine tuning time sync section, but I am unable to prevent this form happening. in fact it keeps getting wost. The only way around the problem right now is to configure windows timesync and then 1000's events are logged in the eventlog of time changes.
Can someone please point me in the direction of what option(s) can somewhat correct the problem? My guests are windows 2008 64bit. The server is a dell T610 with 2 CPU's.
I thought the guest tools should take care for keeping the time in sync, I also enabled the option to sync the time when the service start, that works, but syncing after that of the guest tools does not appear to work.

Thanks

[hvr_za]

the system ran 17 min slow in 12hours.

I almost get the feeling that running virtualbox on solaris/opensolaris is a waste of time. Hardly anyone does it and it seems that the problems are endless. We were thinking about buying solaris 10 entitlements to run virtualbox but it seems that would be a total wast of money. Far too many problems listed of running virtualbox on opensolaris and no resolution, other then comments about, "try running updating your build."

I have slow network issues (had to tool around with MTU sizes etc, and still only at 12 -16MB/s on a 5 disk raidz volume with 7200rpm disks), time sync issues and guest nodes crashing and giving up. virtualbox +ZFS what a pipe dream.

[bqbauer]

I'm not helping, but for what it's worth my guests are keeping sync. I know this because the RTC (real time clock) on my desktop runs really fast. If I don't sync it at least hourly with NTP, it gains at least 1 minute/hour. That's 24 minutes/day! My guests always end up running just as fast if NTP isn't configured in the host, and keep time when the host is correct. This was true with a 2009.06 and b134 host OS, and I always make sure NTP is disabled in the guest.

Unfortunately, this doesn't help solve your problem. The guest additions are supposed to maintain time sync between guest and host and appear to do so for me, so I wonder what's up with yours? Perhaps it's additions with Win2008?

[hvr_za]

Thank you for your response.

If I power-up a system (only tested one of the two nodes so far) on a supermicro server with 2x 1.9GHz HE CPU's I do not have this problem. (same vbox version and sol build.)
Could this be CPU related since it runs fine on AMD but not on Nehalem. Could it be motherboard related or the manner how Opensolaris run on the different CPU types or maybe it is specific to Dell in this case.
Also the response of the VM feels slower on the Dell with 2x2.26GHz and 1067MHz mem than on the supermicro with 400Mhz memory.

[nxnlvz]

This is not going to help either.

Generally speaking I have seen time sync / clock skew issues on all virtualization platforms. VMWare and XEN also have similar issues and is does not matter how you run it. Just go and look it up using any combination of "clock skew" and your choice of product.

One of the most common factors that I see is is any NUMA type processor will exacerbate the problem. This is especially true of the newer AMD processors when all the cores are running at different frequencies. Locking all the cores at the top state will sometimes help but that if defeating the purpose of using those types of CPUs for power saving capabilities. IT kind of irked me the other days when it became apparent that clock skew ( and maybe MTU ) was having a detrimental effect on networking for one of my systems. This one happened to be a 2 CPU quad Core system and until I turned off power management in Opensolaris b111b the clock was acting stupid in a Windows guest and network packets were being delayed and dropped. Disabling the w32time service (you should olny have one source of time) and locking the physical CPU frequencies almost fixes the problem as long as the guest has been started after this has taken place. Of course the what makes me laugh about this is that a clone of the guest running on my desktop that keeps perfect time. The CPU is a slightly older generation AMD without the newest wiz bang feature that allows all the cores to go out of sync which is probably the reason. Of course the newer Intel CPUs have similar traits that allow the same problem to creep in as is evident from my own testing.

All this said. Have you looked at this?
http://support.microsoft.com/kb/835730
and also this
http://technet.microsoft.com/en-us/sysi ... 97568.aspx

What I have seen sometimes is that programs in an effort to have the highest resolution possible will use NtSetTimerResolution to set the interval down to 1ms. I just found an unexpected one today that was doing this. This might be valid for real hardware but does not work on virtualization. What seems to end up happening from my point of view is that the timer interrupts get missed so the whole system gets screwed. I am not sure if there is some way for VirtualBox to tell windows that the minimum TIMER tick can't be 1ms but if it can it should not allow it to go below the 7.8 ms default or more reasonably the 15.625 that is the default for SMP systems. Maybe there is a BIOS flag it could set? i really don't know the possibilities here.

As far as VirtualBox on Solaris. I am using and have seen others use it and will continue to do so. I could use VMWare ESX or ZEN or even OVM but this seems to be what I need right now. It is simple and functional and allows me to piece together what works. I could use Linux as the Host or even Windows. Linux annoys me too much as it has an ever changing ABI and is fragmented beyond imagination. A solution on one thing

[bqbauer]

Yes, I agree. Running VB on (Open)Solaris is not dead, nor is the OS until Oracle says so.

Staying on topic now, I have at at home an Intel quad Q6600, which is the computer with the bad RTC. At work I have an i7-920 (Nehalem). The Nehalem, for me, is keeping good time. The Q6600 doesn't do so without NTP, but on both I do have the CPU set to on-demand versus maximum performance. I'll have to try changing that to see if my home computer alters its timekeeping behavior.

I hope something in this thread somehow helps the original question....

[hvr_za]

All,
Thank you responding.

Before posting I already tried changing the power options of the CPU to OS controlled and then to full power in the BIOS with the same results. Next I will look at the power options of the OS as mentioned.
I will post my results.

Thanks.

[hvr_za]

So far the system has been so stable that I did not get the buy in from the only other person accessing the guest to disable Intel Enhanced Speedstep and reboot the system. THe BIOS setting indicate full power for the CPU's. I will now be making the change this coming Sunday. I will post the result. Keeping time is still a problem, but NTP installed on the guest is correcting time.

[hvr_za]

I disabled all power save settings in power.conf and also ensured non are set in the bios. I also upgraded to 3.2.8. Still, time loss stayed the same.

I will just keep using ntp settings on the host, I really hate the 1000's of events being generated in the evnetlog due to constant time changes.

I guess it really does not matter anymore as OpenSolaris is no more and I should rather focus on what to migrate to.

Thanks for the responses.

[nxnlvz]

OpenSolaris may be no more but Solaris and Solaris Express live on. Good luck on whatever the platform is you select. I encounter timer issues on all platforms and hypervisors. It is always just a matter of getting around the issue.

[hvr_za]

nxnlvz,

Thanks for the response, not that is is thread for this, but Solaris is my 1st choice as I have worked both with Solaris and Linux for the past 10 years. The problem is that Oracle makes it very difficult to buy entitlements for the average person and no COMSTAR in 10. I might stay with Opensolaris until Express released and take a look at what the limitation(s) are compared to a full release.
Please PM me if you have any thoughts around this.
I am really interested as to the route others are/will be taking.

thanks once again for all the input.