yeah, they went over the 'citrix xen' chapter.

But what I was trying to say is that networking under xen is pretty easy for me, as it's just Linux stuff. The times I've had to deal with vmware for clients, I've been frustrated trying to do what I wanted to do with their interface, something obviously designed for windows users. I'm not saying the vmware interface is bad; just that it was designed for a windows user.

'desktop virtualization' which is another animal entirely. (personally, I think desktop virtualization will suffer the same fate of other recent 'thin client' schemes, but really, it's not my area, so that's just wild speculation.)

Provisioning and mass deployments is something I go into a little bit... but certainly isn't the focus of the book. I firmly believe that it is madness to treat physical and virtual servers differently; you want to use one tool for both. And right now, the best one tool is cobbler/koan.

haha. Linode is one of the better providers. they are cheaper than slicehost, but more expensive than I am. According to the latest benchmarks I've seen, though, they beat us all (slicehost, ec2 'small', prgmr.com) in terms of CPU power. (though I'm pretty up front about the fact that I optimize for cheap ram over all else, so the results are unsurprising)

Personally, after the age of 25 or so, I started to agree more and more with Paul Graham on the subject

DomU works, and should continue to work for the foreseeable future; DomU paravirt ops is upstream, so unless debian goes through extra effort to break xen DomU support, it should Just Work.

of the xen stuff- I've had nothing but trouble with it in testing, and looking, it seems that the debian people aren't particularly interested in helping if you have problems, and if you ask the xen mailing lists, they tell you to ask the debian people. If you want stability, you have to deal with the 2.6.18.8-xen kernel distributed by xen.org, or the 2.6.18-patchedtohellandback kernel distributed by RedHat. (the Suse kernel might be stable, I haven't tried it, but the debian port of the suse kernel that is 2.6.27, man, that sucks. Some people say this is because they took the initial patch and have not been good about importing fixes that suse did.)

I've used both those Linux kernels (the xen.org kernel and the RHEL 5 kernel) quite a lot; my experience has been that the RHEL kernel is slightly less stable, while the xen.org kernel has serious driver deficiencies.

For my current production boxes, I'm using the xen.org 2.6.18 system, and I just buy the exact same hardware every time. (Yay for cheap sata_sil cards!)

Another option, of course, is NetBSD5. After starting on NetBSD 3, and switching to linux for pae and x86_64 support, I'm seriously considering switching back, now that NetBSD5 is starting to look good.

Was recommended to me by someone I respect quite a lot, someone who isn't known for being 'fuzzy headed'

But yeah, I read the book and I didn't get it, either.

books can't really compete as references anymore, I don't think. The advantage of a book is that it's easier to sit down and read a book cover to cover than to figure out what you need to look up in order to get an overview of a technology.

hah. being familiar with linux, and not so familiar with Windows, I had the opposite problem. Xen networking is... Linux networking. (the big problem is that the Dom0 kernel is crusty and ancient; something that should be remedied with Xen 4, which should be out Real Soon Now.)

if you pay for power, replacing 8 P4s with 1 dual quad-core xeon or opteron usually saves you enough in power costs to pay for the capital cost in a few months.

For me, the big differentiator is robustness and reliability. KVM will get there, I'm sure. But in the hosting space, Xen is the established tool, and KVM is the new technology. Nearly all the linux developers are rooting hard for KVM, which may mean that it will become better. Personally, I'm hedging my bets by learning both. But for now, at least, I think Xen is the best choice for the hosting provider.

I'm pretty sure I've emailed you to ask dumb questions several times, and you answered. :)

Xen and KVM can do similar things, yes, however, last time I evaluated it (which was about a year ago, and things are changing fast) KVM was wholly unsuitable for production use. The stability and performance had a long way to go before it came anywhere near Xen.

On the other hand, KVM is much easier if what you really need is an accelerated qemu to test something real quick. setting up Xen on the desktop is a huge pain in the ass.

From a service provider perspective, all the parts of xen that need to be upstream are upstream. Guest support has been upstream for some time now. I'm fine running a funny Linux kernel (or a funny NetBSD or OpenSolaris kernel) in the dom0.

"The hypervisor is linux" is the idea behind KVM. The Xen hypervisor is most certainly not Linux.

we've got a chaper on running OpenSolaris and NetBSD Dom0s in the book. We got our first taste of Xen, in fact, on NetBSD before moving to Linux Dom0s (we're talking about moving back, now that NetBSD 5 is out with x86_64 and i386PAE support)

not that I have anything against Persig, but xen/zen puns are pretty worn out. We had a bit of a scare when we noticed that the proof for the rear cover said "Xen and the art of virtualization" - I mean, Our "no persig puns" rule aside, that's what the Cambridge folks called one of the first papers on Xen, so there would have been copyright issues. No starch caught the mistake before it hit paper, though.