KVM under Fedora 9

KVM is a nice virtualisation program. Unlike the basic Qemu it only works on CPUs with virtualisation instructions. Installation used to be a nightmare, but it's now not bad at all.

# yum install qemu-img kvm

Edit /etc/modprobe.conf, adding:

alias /dev/kvm kvm-intel

or kvm-amd, depeding on what is under the hood.

# depmod

Using a disk image we prepared earlier:

# qemu-kvm -hda disk.cow -m 512

Performance of Qemu for boot and shutdown of CentOS 5. Without KVM:

real	2m58.120s
user	2m31.290s
sys	0m4.094s

With KVM:

real     1m8.837s
user    0m27.603s
sys     0m18.641s

As well as the greater elapsed time, look at the proportion of user+sys to real. Qemu alone occupied a CPU for 87% of the time. Qemu with KVM occupied a CPU for 66% of the time. So KVM is about a third more efficient than Qemu alone when both are under high load. This can certainly be felt when using the machines for other tasks as well as running the virtual machine.

KVM is cheap and simple virtualisation, as basic as a Volkswagen Beetle. The sort of thing that's useful when you want to run up a machine to do a test (eg, does my program work on an as-shipped installation) or to build a package for a foreign distribution.

VMWare is better in many ways. Simpler to use, more efficient. But installing VMWare on something other than the supported distributions and versions is a ongoing nightmare of tracking what the kernel folk have changed this week. KVM is in the kernel.org kernel, so that sort of breakage isn't going to happen.

turnitin

These guys are starting to get on my nerves.

Their plagiarism page has a pie chart with this data. Elsewhere on the site they say this data comes from their corpus of plagiarised papers.

How students plagiarize

Entire paper copied	1.00%
Significant plagiarism	29.00%
No plagiarism	70.00%

Now what are the odds of the two data points ending on an exact percentage? One in 10,000. So there's a 9,999 in 10,000 chance that the figures are not exact to degree to which they are presented. That's either sloppiness or academic misconduct (claiming a quality of result not supported by the data).

There is also a category missing. Surely some 0.01% or more of the papers can't be categorised as plagiarised or not?

What are the odds of the convenient 70%/30% split between no/yes plagiarism? Remember this data is claimed to be from real life, not marketing.

A stunning hypocrisy from a site claiming to be opposed to academic misconduct.

Doomsday Book

I read this when it came out then handed it onto the next person. It was reissued with a better cover a few years ago and K bought it a few weeks ago.. Read it overnight (ouch, getting too old for that). I had forgotten just what a stunning work it is: set in today's world (give or take), with a huge amount of imagination, a ripper plot and just stunning writing.

PS: Read the book, not its Wikipedia page. Some Wikidiot thinks giving away the secrets of the plot is a good idea.