Posted by Fraser Campbell
Fri, 27 Apr 2007 03:51:00 GMT
Kernel 2.6.21 was released yesterday, a few of the more interesting highlights:
- VMI now in default kernel (this is VMware’s layer on top of paravirt_ops), theoretically a fully paravirtualized Linux kernel on top of VMware might be do-able now
- updated KVM code which includes some paravirt_ops support and live migration
- clockevents and dynticks
The dynamic clock tick stuff is best covered over at LWN (see Clockevents and dyntick). This could have positive impact on power consumption and performance (particularly in virtualized environments). It will be interesting to see if certain virtual environments can now keep accurate time with a kernel such as this.
Posted in Kernel, VMware, KVM, Paravirtualization | no comments
Posted by Fraser Campbell
Tue, 27 Feb 2007 13:05:00 GMT
Ulrich Drepper is lead maintainer of the Linux C library (GNU libc), works for Red Hat and is well known in Linux circles – basically he is one smart cookie.
Last night he has posted a great article entitled Xensource/VMWare start sandbagging.
The article discusses the fact that Linux already has great support for NUMA, SMP, scheduling, hardware drivers, etc. My favorite “As for better scheduling with a hypervisor: that can only be a joke.”.
We can expect the marketing machines to crank out reams of verbage in the next 6 months but I trust Mr. Drepper more than I would trust anything I see coming from the corporate presses.
Linux as the hypervisor just makes sense. Jeff Dike has been pushing the idea for years with UML and UML continues to move forward.
The only slight beef (or question) I have with KVM is why has it diverged from QEMU? More specifically KVM relies on QEMU for it’s device emulation but it also requires hardware virtualization. Now that QEMU’s kernel accelerator is open source could KVM and QEMU not be merged back together to give us one high performing solution that would support both VT and non-VT capable hardware?
Posted in Xen, XenSource, QEMU, VMware, KVM | 2 comments
Posted by Fraser Campbell
Mon, 26 Feb 2007 20:45:00 GMT
news.com has a very good article about KVM. It discusses plans of some of the major Linux distributors as well as some of the social issues around why KVM is rapidly gaining acceptance while Xen is still fighting for it (within the kernel community).
See KVM steals virtualization spotlight. for full story.
Posted in Xen, KVM | no comments
Posted by Fraser Campbell
Tue, 06 Feb 2007 23:46:00 GMT
With the just announced open sourcing of the QEMU accelerator it would be interesting to have some idea of the performance differences between QEMU, QEMU + accelerator and KVM.
It turns out that someone has already done the hard work for us. See http://linux.inet.hr/ article Finally user-friendly virtualization for Linux.
KVM and QEMU with it’s accelerator both appear to be capable of utilizing CPU and memory at close to native speeds (perhaps 80%). Definitely worth another look on the laptop.
Posted in QEMU, KVM | no comments
Posted by Fraser Campbell
Mon, 05 Feb 2007 00:08:00 GMT
Coinciding with superbowl Sunday is the release of kernel 2.6.20. From Linus’ announcement:
As ICD head analyst Walter Dickweed put it: “Releasing a new kernel on
Superbowl Sunday means that the important ‘pasty white nerd’
constituency finally has something to do while the rest of the country
sits comatose in front of their 65” plasma screens”.
The big features for 2.6.20 (at least for virtualization nerds) are paravirtualization support (x86) and KVM support (x86 and x86_64).
Posted in Kernel, KVM, Paravirtualization | no comments
Posted by Fraser Campbell
Thu, 18 Jan 2007 04:33:00 GMT
linux.com yesterday published an interview with Jeff Dike (author/maintainer of UML). The linux.com interview is available at http://www.linux.com/article.pl?sid=07/01/16/2037250.
It sounds likely that UML will soon be ported to make use of hardware virtualization extensions. This support will be via the KVM kernel subsystem.
Many interesting twists to this story:
- Xen popularizes paravirtualization
- Xen doesn’t get mainline kernel port
- VMware gets on the bandwagon with VMI proposal
- finally paravirt support starts creeping into the kernel
- KVM jumps into the fray (out of the blue)
- KVM now uses paravirt support
- along comes the grand-daddy of Linux virtualization (UML) building on top of both KVM and paravirt
- still no Xen in mainline support
On a related note, Jeff made a very interesting presentation last summer at OLS. The paper was entitled “Linux as a Hypervisor” and can be downloaded as part of the 2006 OLS Proceedings Volume 1.
In this paper Jeff argues that the Linux kernel itself is a very capable virtualization manager. He points out that many of the “fancy” new features being added to the kernel in the past few years have had great benefits for virtualization (async and direct I/O primarily are mentioned).
Another interesting part of the paper was that he mentioned namespace virtualization patches (containers) and how they can be a boon to UML as well by allowing processes to make system calls directly on the host kernel.
One could definitely forsee that hypervisors might be irrelevant before too long. The linux kernel has most of the features that virtualization solutions need (NUMA, SMP, AIO, DIO, volume management, memory management, etc, etc, etc.) why duplicate all of code in a hypervisor as well – UML kernels as a very thin wrapper around the host kernel could yet become the sweetest deal of all.
Posted in User Mode Linux, KVM | no comments
Posted by Fraser Campbell
Thu, 18 Jan 2007 04:32:00 GMT
KVM is definitely generating some buzz these days. Soon it will be in the stock Linux kernel (2.6.20), it is gaining paravirtualization features and in general it seems like a nice, clean, simple solution.
Who would have predicted that the first user of the paravirt_ops kernel functionality is likely not to be Xen or even VMware, it will almost certainly be KVM (dozens of non-mainline patches excepted of course).
LWN posted an good article on these recent developments, author Jonathon Corbet states:
Perhaps the most interesting outcome of all this, however, is how KVM is gaining momentum as the virtualization approach of choice – at least for contemporary and future hardware. One can almost see the interest in Xen (for example) fading; KVM comes across as a much simpler, more maintainable way to support full and paravirtualization. The community seems to be converging on KVM as the low-level virtualization interface; commercial vendors of higher-level products will want to adapt to this interface if they want their products to be supported in the future.
Interesting indeed. LWN’s article includes a fair amount of technical details, it’s definitely worth a read over at http://lwn.net/Articles/216794/.
Posted in KVM | no comments
Posted by Fraser Campbell
Sat, 06 Jan 2007 04:37:00 GMT
In a post to kvm-devel this afternoon, Ingo Molner announced a patch to support paravirtualization under KVM. The support speeds up context switches and TLB flushes.
The email announcement is here, some interesting snippits:
2-task context-switch performance (in microseconds,
lower is better):
native: 1.11
----------------------------------
Qemu: 61.18
KVM upstream: 53.01
KVM trunk: 6.36
KVM trunk+paravirt/cr3: 1.60
KVM will be available in the stock 2.6.20 kernel, great news and at the current pace I’d expect it to be quite a powerful solution before long.
Would be great to hear from Qumranet though …
Posted in KVM | no comments
Posted by Fraser Campbell
Mon, 27 Nov 2006 02:28:00 GMT
Recently Avi Kivity posted to linux-kernel a patchset that adds save/resume support to KVM. See email at http://comments.gmane.org/gmane.linux.kernel/467458.
I have given KVM a run under Ubuntu and while it did work (I ran both CentOS 4 and Windows XP as guests) things were decidedly slow. One would assume that as development progresses on KVM this situation will improve.
KVM would certainly not appear to be the toy variety of virtualization like lhype. Simply considering the founders and backing of KVM one would assume that there is significant functionality and performance to be had from KVM down the road.
Qumranet (the company driving KVM) has as it’s founder Moshe Bar, Moshe is founder of the OpenMosix project and was one of the co-founders of XenSource. One would expect that very interesting things are on their way from this company …
Posted in Kernel, KVM | no comments
Posted by Fraser Campbell
Sat, 11 Nov 2006 15:57:00 GMT
The KVM project website is now up. For those interested in more details see the KVM Whitepaper.
One would wonder if hypervisors are losing their relevancy. With UML already in the kernel and KVM hopefully on the way the kernel already has the native ability to create a “virtual machine” (at least in some sense of the word). The argument against maintaining NUMA, SMP, multi-core support, etc. in both a hypervisor and the kernel makes a lot of sense to me.
KVM is now included in Andrew Morton’s -mm kernels, see announcement for 2.6.19-rc5-mm1.
Posted in Kernel, KVM | no comments
