Virtualization.com

News and insights from the vibrant world of virtualization and cloud computing

Guest Posts

Remus and Kemari , still going strong

by 1 Comment

Remus and Kemari, the 2 VM mirroring solutions for Xen both made some announcements recently.

Remus announced the availability of version 0.9 this release is build against the tip of the xen-unstable repository, and supports PV and HVM in 32-on-32, 64-on-64, 32-on-64, and 64-on-32 configurations. It has been tested using Linux (Ubuntu) PV guests, and both Linux and Windows XP under HVM.

According to the Remus website Last week Remus has been applied to the official Xen repository and is expected to be i
ncluded with the next major release.

In case you forgot Remus provides comprehensive fault tolerance for Xen virtual machines. If the physical machine hosting your VM fails, the backup can take over instantly, as if you had migrated it to the backup at the instant before the failure occured. There’s no need for recovery, because the backup is always completely up to date. Furthermore, Remus runs completely transparently, requiring no changes to your existing guests.

Kemari, the other fault tolerance solution for Xen however announced that the in
itial work for porting Kemari to KVM has started .

The basic design for Kemari for KVM is ready and they have send out a RFC to get early feedback , the full RFC announcement by Fernando Luis Vázquez Cao can be found on the Linux KVM mailing list
More info on Kemari is available on their website

CopyCats in Virtualization

by 3 Comments

Step back in time with me a couple of years, Xen was starting it’s upmars and with Xen paravirtualization became popular , then came the other Virtualization vendors and the discussion about which technology was best discussed started.

Now we all know that VMWare and Xensource were discussing how to include hooks for Paravirtualization into the Linux kernel and eventually that also happened,however there wasn’t really any adoption , a couple of weeks ago VMWare announced it was going to drop support for paravirtualization. Aparrently VMWare’s Paravirtualization story wasn’t really a success.

The Xen folks pioneered with Paravirtualization and a dedicated hypervisor, yet somehow also got interrested in running the Xen engine within an already existing operating system by means of a kernel module. That way Xen can also be run on different existing platforms just as KVM and Virtualbox, one of the big reasons for KVM adoption exactly is the fact that one can turn an existing Linux machine into a virtual machine host by doing a simple modprobe

However the initial development focus for Hosted Xen was Windows and OS/X
not really a market wher Open Source Virtualization tools are going to make big
adoption steps fast

So you might wonder who is using this Hosted Xen anyhow , apparently not that many people. On the other side there’s VMWare’s ParaVirt similar story
And who’s using ParaVirt ? Aparently nobody …as there used to be a huge performance bug in it for over a year ..

So the lesson learned ? Stick with your own mainline technology .. no need to
copy the others ideas, seems like they won’t be a success anyhow ..

Project XVP

by 1 Comment

On blog.xen.org we read about XVP , one of the most common questions for XenServer is a webinterface to manage the VM’s from a browser.  XVP , developed by Colin Dean at the Durham University in the UK has one based on 4 components.

  • xvpweb: A web front end for XenServer, running on Apache/PHP. It will provid
    e you with status views , facilities to shutdown, reboot and look at virtual mac
    hines and their consoles
  • xvpviewer: A Java applet for accessing XenServer consoles (used by xvpweb, of course). a custom Java based VNC client, based on TightVNC but with xvp specif
    ic additions.
  • xvp: A server-side proxy for XenServer consoles, so that you can use ordinary VNC clients xvp stands for Xen VNC Proxy it will be used as a VNC proxy to the
    consoles hosted on a Citrix XenServer platform
  • xvpdiscover: A tool that queries a XenServer pool, and writes the appropriate configuration files for xvp and xvpweb.

Screenshots and rpm’s for RHEL based distributions are available from their website

Enomaly ECP 3.0.3 Service Provider Edition Released

by 2 Comments

Reuven Cohen, Founder and Chief Technologist for Toronto based Enomaly Inc posts on his blog Elastic Vapor that Enomaly is announcing the general availability of The Enomaly Elastic Computing Platform (ECP) Service Provider Edition 3.0.3.

According to him this new release brings the following new features :

– KVM is now directly supported as a hypervisor at install time.
– Sample data is installed during initial installation, so there is no need to create a customer/group/permissions before testing the system. See INSTALL for default user/pass.
– VNC window in customer UI is now identical to the Admin UI. Passwords for the VNC console are now found under Info button at VM level.
– Info window now shows how to connect with an external VNC client as well as the existing Java applet.
– VNC window can be disabled entirely on a per VM basis.
– App Center can now be searched/filtered. This is useful if you offer a large number of appliances.
– Admin Dashboard now shows graphical whole cluster resource usage.
– Network Manager has been removed. All deployments are recommended to use DHCP for IP assignment going forward.
– Various performance improvements have been added at customer UI level.
– Various performance improvements have been added to infrastructure code.

Memory OverCommit and Shared Memory pages

by 1 Comment

Have you ever been in that meeting with the proprietary sales guy on one side and a bunch of Virtualization newbies on the other side ?

Usually the simplified discussion goes like this,

Customer : “We’re looking into virtualizing some of our servers , we already have some windows, and some Linux, Linux does Virtualization right ?”

SalesDroid, realizing he is going to miss out some license comission if he does’t spread some FUD quickly : “Oh great, we have some awesome products to deal with that”

Customer : “But can’t we allready do that with the RHEL/SLES we have ?”

SalesDroid: “Oh no, those platforms have some basic virtualization functionalitym, but they don’t have support for Windows, Live Migration, Memory Overcommit, they don’t have support Storage, no VLAN support etc”

Customer : “Oh we didn’t realize that… can you make us a quote for your product ?”

So you bump into these folks in the pub and they tell you why they opted for the commercial product, and you go like “What moron told you that” , Xen, KVM have all the features you just described ..

Afterall Xen Memory Overcommit is something you could replace with a small shellscript
but as of recently KSM is available in the standard Linux Kernel. As KVM Virtual Machine guests are just processes under linux that means that you can use this stock Linux feature to have your Virtual Machines benefit from the feature

In fact by having multiple processes share the same memory you create a memory overcommit feature for KVM so that it can make better use of the available memory. A according to Linux-KVM RedHat claims it has tested KSM with 600 Virtual Machines running on host with 48 cores and 256Gb Ram.

So now also KVM can tout this feature 😉 Hower the discussion stays if overcommit is the right thing to do, afterall there is it’s not a one size fits all discussion. Overcommitment cost performance, you’ll eventually end up swapping which is painful, and the last thing we want to see is the Out Of Memory Killer kick in action , so maybe it is better to cough up some more money for the memory afterall

The End of Paravirtualization?

by Leave a Comment

Alok Kataria , a Systems Programmer & Linux Kernel hacker at VMWare posted to the Linux Kernel Mailing List last month about his experiences benchmarking VMware’s paravirtualization technique (VMI) and hardware MMU technologies (HWMMU) on VMware’s hypervisor.

Their test platform was a Dual Quad Core AMD Opteron 2384 2.7GHz (Shanghai C2), RVI capable running a development build of ESX.

They ran different real world test on a SLES 10 SP2 with a 2.6.16.60-0.37_f594963d-vmipae Kernel and they concluded that in most of their benchmarks the EPT/NPT (hwmmu) setup provides equal or provide better performance compared to VMI.

Below is a short summary of performance results between HWMMU and VMI.
These results are averaged over 9 runs. The memory was sized at 512MB
per VCPU in all experiments.
For the ratio results comparing hwmmu technologies to vmi, higher than 1
means hwmmu is better than vmi.

compile workloads – 4-way : 1.02, i.e. about 2% better.
compile workloads – 8-way : 1.14, i,e. 14% better.
oracle swingbench – 4-way (small pages) : 1.34, i.e. 34% better.
oracle swingbench – 4-way (large pages) : 1.03, i.e. 3% better.
specjbb (large pages) : 0.99, i.e. 1% degradation.

In his post Alok tells us that VMware expects that these hardware virtualization features will be generally available in all hardware by 2011. Other reasons for VMWare Paravirtualization’s VMI , such as timekeeping have also left the building therefore VMWare decided that they plan to stop supporting VMI

In short … VMWare has prepared a patch to remove all VMI related code from the Linux Kernel.

The thread on LKML went on noting that current users still might want to have the patches present so immediate removal of the feature wasn’t really the best way to proceed. But some day soon the code will be removed from the standard Linux kernel.

So what does this say about the future of Paravirtualization ?