virtualbox.org

I have a Windows 7 guest running on a RedHat 5 host. When first installed, Windows was very sluggish. I turned off Aero and all visualization enhancements and that seemed to fix the problem. But, it's now sluggish again, and all of those enhancements are still off. When I say it's sluggish, I'm referring to the display. Actual processing is as fast as it should be; it's just the display that is very slow.

The computer itself is a Dell Poweredge 2900 with 2 Xeon quad-core processors,and 32GB RAM. I have allocated 6 cores and 16GB to the VM (it's for scientific data processing, so I need a lot of horsepower in the VM). I wonder, however, if the issue has something to do with the graphics system. Despite the power of the rest of the machine, the display adapter is only an ATI ES1000 with 16MB memory. Even in the host OS (RedHat), the graphics performance leaves something to be desired.

Any ideas what might be causing it to be so sluggish and what I can do to fix it?

Without any graphics acceleration you should not need more than 16MB of vRam. Remember that the vRam of a guest is actually using the system RAM and not the memory of the GPU.

I know the common rationality to get more out of a virtual guest is to slam more processors at it, but I can tell you from experience that this is not always the case. Sometimes it works better with one or two. If I were trying to isolate this I would start with the processor count. Then I would make sure that the Windows guest was de-fragmented and double check for unneeded processes.

I have allocated 18MB vRAM because of the logic you stated. I would go with less, but VirtualBox complains with any less than 18.

On your advice, I tried reducing the core count to see if that makes any difference. That kind of freaked Windows out, and it's trying to repair itself at the moment. I kind of hope that doesn't solve it, or if it does, it will point to a way to fix it and keep the 6 cores. The only reason this VM exists is because of one Windows program I need to run on the machine. Processing the data this software is for can take many hours, so if I can only use one core, that will really suck.

the 18MB vRam is fine. as for the amount of cores like I said you would need to tweak it and find your right spot. I have had 2 work almost 50% faster than 4 on several occasions. Here again this may or may not be your solution but this has really been a factor in some of the guest (particularly Windows) I have worked on.

Okay, I just finished trying a few things. I tried using several different numbers of cores, from 1 to 6. First of all, there was no difference in the general sluggishness of the interface. Graphically, it made no difference; windows still scrolled just as slow, etc. However, I did notice a dramatic difference in the time it took to boot and shutdown the system. With only one core, it took about 25 seconds to boot up and 4 seconds to shut down. Those times increased with more cores. With six cores, it took just shy of 3 minutes to boot up and nearly 30 seconds to shut down. I tried processing a small amount of data, and with one core it took 16 seconds and with six cores it took 21 seconds. There's definitely something odd going on there. Any ideas of things I can change to make it actually faster with more cores? If not, I may be forced to try a different virtual machine emulator because I really need to be able to utilize the power of this machine.

Probably not with a type II hypervisor. At Least from what I have seen. As you see reducing does increase speed (sometimes anyway) and you need to find a spot that will do what you want. I would leave it set to the fastest and let it do its thing for a day or so. See if it is as fast or faster doing the big job as it was before. Also don't forget to check for fragmentation as I suggested before. Windows has a tendency to mangle things a bit when it compiles and needs to be restructured every once in a while to put things back in order.

Just to be clear VirtualBox is not an emulator. These would actually be a lot slower. You might ask the DEVs on the DEV mailing list or IRC (probably faster answer) what they think. As for the speed, VirtualBox comes in second only to a type I hypervisors in almost all benchmark tests I have seen, and in every test I have performed with other type II hypervisors it is fastest.

Like I said common rationality is to add cores but it does not always work that way in real world situations in a virtual environment.

https://www.virtualbox.org/wiki/Community

After running some benchmarks, it seems the CPU scores increase with more cores, but the 2D graphics scores, which relate to the usability issues I'm having, decrease somewhat with more cores. (see below for more details) If possible, I would like to fix this interface sluggishness problem. So I have some questions about how VirtualBox works, specifically with regard to the graphics system. Even in the the host OS, RedHat 5.7, the system seems somewhat sluggish. Scrolling through documents or webpages is jerky, moving windows is jerky, and there is always a little lag between user input and what happens on the screen. It's not a lot, but enough to notice. In the VM, it's the same, just a lot worse. I believe the issue in the host OS is due to crappy drivers for the GPU (ATI ES1000, a low power GPU used in servers). I booted the system into Fedora 16, a much more recent OS, off of a USB flash drive, and the system was very snappy and exhibited none of the sluggishness seen in RedHat, so I know the GPU is capable of a quick response. Does VirtualBox use the host GPU at all, or is the GPU entirely virtualized? Could the poor graphic performance of the guest OS be related to that of the host OS? If the problem is the GPU, then a simple solution is to just get a new video card with better support in Linux, but if that's not the problem, then I need to fix something in Windows or VirtualBox. I would prefer to just replace RedHat altogether with something else, but that's not really an option, unfortunately.



This isn't really useful in solving my problem, but I'll post this information for anyone reading it who might be interested: I've done a little bit of testing of this system. I installed some benchmarking software and ran a bunch of benchmarks using different numbers of cores. The summary is that for CPU tests, the scores scale very well with the number of cores; the more cores, the higher the score. I graphed the results, and the lines are just about straight in all cases except one. The physics CPU test was a little weird in that it started at 83 with one core, increased to 347 with 4 cores, then dropped to 124 with 5 cores, 3.8 with 6 cores, and 1.3 with 7 cores. The disk and memory benchmarks were pretty flat and didn't really change with the number of cores. The 2D graphics benchmarks generally stayed flat or decreased with increased numbers of cores. Of those that did decrease, the 7-core score was about 60-70% of the 1-core score, so those decreases weren't huge. I also timed the startup and shutdown times, and saw dramatic increases with more cores. With one core, the startup and shutdown times were 23 seconds and 4 seconds, respectively, while with 7 cores, 195 seconds and 24 seconds. The lines on those graphs were pretty straight as well.