Will Red Hat come back to haunt the Open Virtualization Alliance? In an attempt to cripple VMware's lead in enterprise virtualization, a posse of tech leaders decided to band together behind Red Hat's KVM in the Open Virtualization Alliance (OVA), announced this Wednesday. The irony, however, is that these same companies may come to rue their propping up of an already strong competitor, one …
In many ways, virtualisation in the x86 market is becoming commoditised. We have Amazon EC2 based on Xen - though for most punters that doesn't matter: it's just a bunch of x86 capacity at the end of a bit of wire. Oracle are also backing Xen with OVM. We have Microsoft lining up a range of vendors behind Hyper-v cloud, with the basic Hyper-v offering being free (gratis), and a public cloud offering (Azure) which again is a bunch of x86 capacity at the end of a bit of wire.
Red Hat may be keen to build up an ecosystem behind KVM, but we're past the point where the virtualisation engine is a differentiator: it is the overall experience which matters.
The reason to bother, is that RedHat are offering an open stack, while vmware tout a proprietary one...
It's much easier to compete with open, as it's much easier for a competitor to provide a migration strategy. Migration away from a proprietary stack is often much harder, by design.
Also KVM is not really a RedHat technology, it is part of the standard Linux kernel...
My biggest gripe with VMWare is that you are tied into their proprietary management tools, i would much prefer to be able to manage my virtual machines using standard tools like a browser, ssh client and vnc. I certainly don't want to be forced to keep a windows machine around to run their proprietary binary applications, and be prevented from managing my virtual machines from new devices such as tablets.
RHEV-M still runs on Windows ... sure, they've almost(?) finished creating a Linux version, but it's always been a point of amusement for me that RH's virtualisation solution requires a Windows instance to manage it.
For the uninitiated, RHEV-M is a proprietary Management system that RH sells, KVM doesn't require it by any means.
> They have roughly the same price tags
Only if you pay for it.
There are assorted RHEL rebuilds that offer KVM without paying a penny. They work just as well as the RH solution - but you don't get the RH support guys on the end of a phone line.
Vic.
[Disclosure: I rebuild and ship RHEL source to my customers]
Whats wrong with the vSphere PowerCLI and SDK for Perl? I do a lot of Host and VM management via cli on remote linux servers. VM backups using the VIBE perl script from NetApp work nicely (rather then having to pay for the crazy expensive NetApp commercial package). You can also configure the VM's to allow you to connect via VNC rather then having to use the vSphere client. Of course, I just set up a Windows VM to run the vCenter server, and run the client through that when lazy.
A poor hypervisor, through either poor software or bad installation or configuration is as bad as being badly managed. Poor performance or reliability have jaded a good number of people against virtualization. Management tools are meaningless if the underlying hyperviser doesn't function as it should.
MS particularly peddle the line that the hypervisor is just a role of the OS. Such a lax attitude explains why it's a clunky, unreliable kludge strapped to Windows that they essentially have to give away 'free' to build market share. The management stack is it's only redeeeming feature, and it too is overly complex for what it aims to do.
Gartner can get cold hard numbers from VMware and extrapolate that over a sample set to get the number of instances in your average hypervisor cluster. The "cold hard numbers" it might get from RH aren't representative of KVM's installed base (in fact, many organisations don't use RHEL but instead one of its many open source derivatives), and then will use it differently to how they'd use VMware. But then again, maybe they're right....
Virtualisation is often a solution to a problem that many Linux shops don't have... partitioning hardware because they have software that doesn't scale. Virtualisation is often used only for R&D and QA environments, where performance is of less importance and speed and flexibility of provisioning is key. It may not ever touch a production box.
If you can deploy well designed software that scales across CPUs and can run alongside other software on raw hardware, then you've got no need for virtualisation.
In my organisation, we do use KVM extensively in production for more than one reason (we have enough legacy software that doesn't scale and/or won't guarantee to play nicely with other apps; we also have multiple independant operations teams which we want to slowly merge; and we have problematic network and DC management (a more fundamental and definitely the most frustrating problem) and we can reduce the pain with virtualisation). It's my firm belief that we should be minimising the use of virtualisation purely for performance reasons. As an aside, I'm keeping a very close eye on LXC and OpenVZ (as they port to cgroups) for lightweight virtualisation alternatives. My kingdom for Solaris Containers!
Having software that can scale and share the same instance (or not) still leaves the unanswered question.. what's your platform? Virtualisation doesn't give you a platform.. it gives you infrastructure.. How do you install and configure your OS correctly. By hand, with a checklist? Network-install with post-install scripts? Puppet/CfEngine? Hosting services? How do you deploy new releases of software? And if rollback is required? Multi-Datacenter? How do you abstract away the differences? How do you get your datasets from A to B reliably and fast? Will your applications survive a DC failure?
While virtualisation is an important and liberating technology, Platform was, is, and will always be the true differentiator... and almost all non-trivial solutions are developed in-house, and most are never shared.
Red Hat would probably disagree. Licenses are where the money is, and support's a nice extra to throw in (helps with the sales pitch of selling free/'free' software).
A potentially unlimited number of licenses for a limited amount of work can definitely be more profitable and is definitely more scalable than a limited number of support staff & consultants who are only able to serve a very limited number of customers.
Now, you do have enough profitable body-shops out there, but they make their money by sending 10 average-joes for every half-decent engineer they have, and then throw in a project manager or two for extra profit, and are usually charging per-person 4-6x what a good employee would cost (not taking into account the extra baggage of PMs).
Red Hat isn't one of these companies.
Kvm is virtually invisible
Xen is a lot bigger and proven to work. And free, if you want to. Same for virtualbox
Virtualbox ose is bigger in share, better and for server use you dont neeed the closed stuff
kvm equals poor performance and scalability, just run a few days stress on itandyou.will see it collapse......
ac
"All of which leaves us with the question: why bother?"
All those executives gotta look like they're doing something...
The top guy complains about VM ware, a bunch of midlevels spend a few month researching, writing opinion papers, then agreement papers, etc...
They get be out of the office a load of times to party with their collegues from the other corporations and a perfect excuse for a nice big travel expense sheet.
And because they're sooo busy on a strategic deal like this, their job is safe for at least till the plan fails. Since its a faith based thing, a failure will be easy to explain and has a less than 20% chance to get them axed.
Excellent all around.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
