Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I've got a Windows 2003 server (64-bit) running as a VM on a remote hosting facility. (I'm just leasing this one particular virtual instance, so I don't know what sort of underlying hardware it's running on, other than that it's presenting itself to the VM as having 8 CPU's available.)

The problem is that starting about 1-2 weeks ago, taskmgr.exe began showing something like a 60% total CPU load, spread out evenly across 7 of its 8 procs, but with one proc spiked at 100%. And the server is responding like you'd expect when it's that busy: it's a dog. I'd obviously like to track down what's causing this.

The problem is that the CPU %'s for each process, as shown in either taskmgr.exe or procexp.exe, don't add up to anywhere near 100%. In other words, the system idle process is somewhere around 40%, and a few other processes maybe add up another 10%, but where's the other 50% coming from? In other words, something is chewing up 50% of my CPU, and it's not listed anywhere in task manager. ("Show processes from all users" is checked.)

I've tried stopping all the services I could, but none of them had an impact on the CPU. Restarting the server doesn't make any difference: by the time I log back in, the CPU is pegged again. Procexp.exe doesn't show anything out of the ordinary.

I can think of two possible explanations: (1) There's some sort of rootkit that's made its way onto my server and is hiding itself from the process list; or (2) taskmgr.exe is suddenly (and for the first time) showing utilization on the rest of the box, not just this particular instance (though that doesn't seem right).

Any other suggestions for tracking this down?

share|improve this question

1 Answer 1

up vote 1 down vote accepted

I see two possible things you should look into.

First, whenever I hear someone talk about high CPU load, without being able to identify any offending processes, IO contention is my first guess. When there is high IO contention, processes stack up in uninterruptible sleep state, filling the OS's process scheduler with tasks that are just there waiting for data to be read from or written to disk. The individual processes would not show as having high CPU load. You'll need to look at performance statistics for the disk subsystem that's servicing this VM to see if any one or more of the many possible IO bottlenecks are being hit.

Second, you mentioned having an 8-CPU VM. Are you absolutely sure you need that many cores? You're sure? Okay, ask yourself again if you really need them. The reason being is that under virtualization, multiple cores don't work the same as if you were running on bare metal. The only time your VM is going to get CPU time on the host is when there are 8 cores available. If 8 cores aren't available, you don't get CPU cycles. On an even moderately-loaded host, needless to say, it's much more difficult for the hypervisor to schedule CPU time for an 8-core VM than it is for a single-core VM. For this reason, I recommend sticking with a single core unless it's 100% absolutely necessary for the application, at which point I may allocate 2 or at the very most 4 cores, in which case I'll make darn sure that there's not much else going on on the host, so that this VM's performance doesn't suffer.

So - may you have a rootkit? Sure, possibly so, and you had better do some due diligence to determine whether or not that's the case. If not, though, you certainly have some other things to look at.

share|improve this answer
    
Thanks for your suggestions. Unfortunately, I'm not sure they apply in this particular case. Even if disk IO on the host's disk subsystem was a problem, or if it were more effective to use a single CPU, I don't have any services running that should be affected by those problems. This is just a test server, with nothing happening on it at all except when I'm running a specific test. Plus, since I'm just leasing this VM from 1&1 hosting, I don't think I can control the number of CPU's presented anyway. I'm running a virus scan on it now to see if there's any kind of rootkit... –  Ken Smith May 20 '11 at 5:21
    
What hypervisor is this running on? –  EEAA May 20 '11 at 5:22
    
I honestly don't know. The VM instance comes with Parallels Plesk management console installed, so they might be using one of the Parallels products for the virtualization as well, but that's just a guess. –  Ken Smith May 20 '11 at 6:49
2  
This might be a memory overcommitment problem on the hoster's side. I have seen similar symptoms (CPU usage for trivial processes like taskman spiking to unbelievable values) when the hypervisor started swapping. –  the-wabbit May 20 '11 at 7:33
1  
I suspect an overcommit problem (possibly memory, possibly CPU) as well. Vmware 3.5 and I believe parallels have to wait for a free core to schedule time on. vmware 4.0+ and hyper-v have a differenet method of co-scheduling that does not cause contention with mutiple core machines. As a hosting provider they shouldn't be using memory overcommitment, so if they are I would suggest finding a new provider. –  Jim B May 20 '11 at 13:26

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.