This has happened a few times in the last three years. Hence we are sure of this.
Because of dust buildup the CPU or GPU in some notebook computers gets hotter than usual. E.g. the GPU could be around 72°C instead of around 59°C. You may hear the fan spin faster and louder.
Simple software runs just fine. But…
if you are doing something which uses the processing units, e.g. watching a video stream, everything on the computer becomes horribly slow, really slow, behaves as if less than 10% of the processor is all it has, almost not workable, video stutters, other virtual machines are very slow, may take a minute to close a window, and takes a long time to recover even if you finally can stop the video playback or close one virtual machine.
This apparently isn’t just the effect of the processor protecting itself from overheating, and slowing down from 2.5GHz to 1.6GHz.
We have observed it on Enterprise Linux 6.x to 6.3 versions hosting VMware Workstation 7.x to 9.0.1 versions running Windows XP SP3 virtual machines with current VMware Tools installed.
Last week it has happened, again, on a high-end Dell Precision M6400 with quad-core and an Nvidia Quadro FX 3700M.
The problem went away when using a vacuum cleaner to suck dust out of the notebook’s air intake slots at the bottom of the notebook computer, both on the left and on the right, while the computer was running.
Also, although sucking the other direction, dust clouds blew out of the back of the notebook, which was easier to see against a dark background.
A temperature drop of 10°C and more has been observed every time we fixed this problem this way, possibly starting with a 5°C drop within seconds and then a minute or longer until realizing the full drop.
Evidence and theories
We do remember the problem begins to be noticeable at the same time as inside Windows XP virtual machines Process Explorer shows spans of time with high CPU use of System Interrupts (Hardware Interrupts and DPCs). At the same time in the host Linux System Monitor there is increased CPU use for vmware-vmx instances, even if they’re not doing much. Possibly there is high CPU use for gnome-system-monitor itself.
A good starting point for explanations is, when the GPU (or CPU) notices it is getting hot it runs slower. One may assume, some (software) component doesn’t play nice, and instead of smooth gradual degradation things get really bad. One theory is, one (software) component doesn’t recognize a slowdown due to overheating (e.g. VMware might not have tested for that, but we can’t say they didn’t), and because of unexpected timing deviations does something that ends up being more resource consuming than it should be, worsening the situation.
We cannot say for sure whether it has to do with video playback, or whether it has to do with VMware. We cannot say for sure whether it occurs without VMware, because we use VMware that much. We cannot say for sure whether it occurs on Linux hosting VMware only. Too many machines throughout the years, just don’t remember.
Couple more thoughts
It would take weeks for the problem to recur. That may depend on the dust in your environment.
We also have changed settings for notebook GPUs to run slower, but reducing GPU speed apparently drops the temperature less than vacuum cleaning does, maybe 5°C.
You got to read this story after the problem has been solved. We, however, first suspected many different causes: Bad new versions of a graphics driver, the host operating system, the guest operating system, VMware, malware, a hardware defect, etc.
We can imagine people buying new a notebook computer because they think theirs has become too slow to run modern software, when the real problem has been dust.