Slashdot Mirror
Inside VMware's 'Virtual Datacenter OS'
snydeq writes "Neil McAllister cuts through VMware's marketing hype to examine the potential impact of VMware's newly pronounced 'virtual datacenter OS' — which the company has touted as the death knell for the traditional OS. Literally an operating system for the virtual datacenter, VDC OS is an umbrella concept to build services and APIs that make it easier to provision and allocate resources for apps in an abstract way. Under the system, McAllister writes, apps are reduced to 'application workloads' tailored through vApp, a tool that will allow developers to 'encapsulate the entire app infrastructure in a single bundle — servers and all.' The concept could help solve the current bugbear of programming, parallel processing, McAllister concludes, assuming VMware succeeds."
So it's not just vaporware, it's an "umbrella concept" that will be built into future products.
No folly is more costly than the folly of intolerant idealism. - Winston Churchill
The real nut of my questions is, what would we need to add to more conventional OS's (linux) to get the job done? For my money, the biggest problem is package interdependencies. IMHO much VM usage is actually just to address that issue. We need package management that isolates applications from each other, giving the appearance of a custom chroot environment for each, while silently sharing resources (such as .so's) that just happen to be the same in multiple applications.
FTFA: "In short, if done properly, a meta-operating system based on networked virtual machines could streamline software development, make IT more flexible, and save customers money."
It is hard to argue with a truism. But what does "done properly" entail?
Getting traditional "silo" orientated programmers to use distributed computing is hard now!
This server is for chocolate, this one for peanut butter... don't let them touch!
Even GRID enabled software like Informatica is hard to get them to understand. Don't worry where it runs, don't try to segregate workloads... the software is smarter than you!
Let it do it's damn job.
Never answer an anonymous letter. - Yogi Berra
"Pronounced"...
I do not think it means what you think it means.
there is hardware available with XEN, that does just that concept. Of course, it is Linux in there, but each major app has its own set-up. That way, you have a DB, a webserver, a development env, etc.
I prefer the "u" in honour as it seems to be missing these days.
I have used VDC OS. Ultimately it is just a convergence of the existing technologies Vmware has already been developing, upgraded to a new level. I can say, it is very, very nice and clean.
What it gives a data center manager is abstraction and ease of use. The physical way everything is deployed one-off into a datacenter, you need a new application, it involves buying new servers, racks, power and whatnot. If you need to move those servers to another center, or deal with business continuance and disaster recovery, it is a new discrete project.
With VDC, no more. You build all of that into the datacenter "OS", and when a new application comes along they are put into the VDC OS and they inherit everything, not just HA but BC, DR and all of the ease of use. If they don't want BC or DR, they don't pay into that bucket.
Need to move a Datacenter? Use the DR solutions in VDC OS, and you can do it in the middle of the day without your users noticing more than a slight 5-minute bump (or so--largely to let the network routes update).
VMware is so far beyond everybody else in the virutalization industry, it is almost comical to hear other people shout the battle cry of 'Xen' or 'Hyper-V'. Those are nice toys, but the surrounding tools are klunky and almost non-functional, leaving just the hypervisor. What VMware is trying to say with "VDC OS" is that the game already left the hypervisor, that is why everybody is all but giving the hypervisor away for free now.
I may sound like a fanboy, but after having worked in the datacenter for 15+ years I can say this technology really works, and its about time. We can now move the datacenter from the hobbiest market it has been in up to now, into the dialtone it should be.
Remember. If this works, NOONE will prefer to keep their own data, their own apps and programs... everyone will ENJOY going back to the days of just using a dumb terminal. Speaking as probably one of the youngest people to have had to use a green lined plaintext terminal from a remote location back when we were moving and my dad had to keep the home computer up... I think I'll stick with having my services local. Nothing worse than not being able to play nethack because the internet wire is down...
Never disregard the raw power inherent to stupidity... they call it "dumb luck" for a reason...
Except from IBM of course.
vmware is simply the logical extension of what the OS should be doing anyway.
or put another way.
Those who don't buy IBM kit are condemned to reimpliment (badly, and for the rest of their lives) what IBM have been doing for decades.
Deleted
We have IEEE and RFC for standardization of ethernet/switching and routing respectively. What standards exist for virtual environments? As commercial security vendors move into this space, we're headed back into a cycle of supporting multiple architectures. "Security Vendor X" must now understand how VMWare, Hyper-V, Xen, and other VM environments perform their networking. Virtualization of the entire OSI model renders the physical and data link layers obsolete. Why emulate them at that point? Not to say ethernet will disappear, but I can see a point where operating systems evolve branches that run in pure play virtual environments. Those offshoots will shed unnecessary things like MAC addresses as the VM vendors begin defining the new network standards themselves.
Openmosix project closed earlier this year and suddenly vmware has a way to run one "OS" over multiple computers. Hmmm...
VM? LPAR? Parallel Sysplex? Haven't IBM mainframes been doing this since the '70s (okay, Parallel Sysplex has only been since the '90s)?
No doubt a "cloud" of UNIX boxes is harder to marshall than a couple of zSeries though.
Theres an interesting little article on the future of these virtual data centers and the kinds of management features/software that will be required to drive them over at the 360is blog. They draw parallels with the evolution of storage management and NAS/SAN features we have grown to love over the last 10 years.
AG.
I fail to see how this "solves the parallel programming problem". If you have a monster server, bandwidth and latency are low for process running on it and communicating with one-another, whether they are running in a VM or not. If you have the same server running *nix with all the programs running, the performance can't really be worse than if you use it to host this OS. It would just be harder to maintain. The only useful feature I saw from the article is that it seems to be able to checkpoint guest OSs. They can't be the first to offer this, not counting IBM, but they might be the first to have it integrated and support un-modified guest OSs.
Disclaimer: I have a CS degree, but I've never used VMware other than Server and Workstation
And (for many of them) it's never going to get any easier.
It is too easy for them to just think of "one program, one OS, one machine".
Their app takes all the resources it sees from the OS it sees on the machine it sees.
So VMWare "solves" this by making it easy (for a price) for each app to believe that it has it's own machine. So the programmers can keep working they've always worked.
As a dinosaur who started cranking code more than 40 years ago, I've been out of touch with things like virtualization for some time. The last word on virtualization in the mainframe world in the 70's and 80s was IBMs VM series of virtual machine operating environments: CP/67, VM/370, VM/SP, etc., coupled with CMS, the Conversational Monitoring System OS for each virtual machine. These were spectacularly useful across a wide spectrum of user profiles. In concept, how do current virtualization strategies differ from these early IBM products, and what took so long for products of this type to return to the mainstream?
VMWare is neat and has its uses. As a developer, I've found it quite useful for OS development and testing. In the data centre it too can have its uses, but it is also has its limitations. That's one of the reasons why our IT department is exploring the Trigence solution--application virtualization. It gives them better performancee, easier migration of apps to newer OS versions and lower costs (hardware and fewer OS images to maintain).
thhis isn't new, amazon's been doing it for a long time
I understand and believe that moving VMs around works well for machines serving stateless transactions, but I'm a tad skeptical that they can move a machine and keep all of its TCP connections alive during the migration. Is this problem solved?
And after a few years when Microsoft follows VMWare, we'll have Microsoft DataCenter OS, abbreviated MS-DOS.
This is a sticking plaster for the lousy PC architecture which today is being forced into places it was never designed for. Read up on what IBM and Tandem were doing back in the 70s and 80s which hardware that was designed for this.
This isn't hot new tech, its putting lipstick on a turd so companies can save a few pennies.
But people in IT rarely read up on their own history so think everything they haven't seen before is cutting edge tech.
Mainframes in the late 80s suddenly became big , nasty and old fashioned systems and desperately untrendy. The PC and unix boxes suddenly became the system de jour and all the supposed hot new talent went in that direction. Unfortunately , not being very good at reading history they had zero clue as to what mainframes actually got up to and so its taken them this long to effectively re-invent the wheel. So endeth this tale.
I think it is, actually.
We've got some VMware guys at my job doing a proof of concept for us. (I work for one of those big companies where people hear the name and that cha-ching noise happens in their head.)
Each VM has its own MAC address, and the virtualization layer includes a network switch. So long as the switch knows where to send the packets, and the other end of the TCP connection is willing to tolerate a few moments of silence while the VM moves, it should work.
The preferred solution is to not have a problem.
Yep - the "cutover" happens faster that the TCP timeout window. The connection stays alive, and even if a packet is lost it simply gets resent when the ACK goes missing.
...a universe of sandboxes.
Orwell: "In a Time of Universal Deceit, telling the Truth is a Revolutionary Act"
Yes, VMware provides a technology for its datacenter level products called VMotion that does exactly that, moving VM's between physical virtualised servers in a cluster while preserving all active networking connections.
I don't know the specifics of how it works and manages that feat, but I have seen tech demos that show it in action. I watched one a while ago published by Dell showing a VMotion task in progress, I'm sure you can find it on the web somewhere with some digging around.
Regardless, it does work and has been available as part of its enterprise products for some time.
If you want to know more:
http://www.vmware.com/products/vi/vc/vmotion.html
Disclaimer: I work for VMware, and I just came back from VMworld in Vegas (exhausted BTW).
In all my 5 years in the company, I must say that this is the most comprehensive re-thinking of the long-term strategy for virtualization I've seen to date. It brings a new sense of direction that matches where the markets are going.
I agree with most of the comments in this thread regarding the benefits of the VDC-OC, but this is just one part of this picture. IMHO, the biggest change is the "Federation with the Cloud" strategy, where a company may choose to use, move or spawn new or existing workloads directly into a service provider on-demand, maintaining the SLAs (from security to capacity) and then bring them back to the internal cloud if needed.
I mean, go a talk to a CFO or a COO, and they'll [most of the time] politely complain about IT being expensive, and not fast enough to react to the changes the company needs. Shared services are still seen as optional and many business units still prefer to implement their own thing. With this model, IT becomes a true utility, with a pay-as-you-go menu that implements a coherent chargeback model that will bring a smile to the guys in dark suits.
Even if VMware doesn't succeed in these efforts, the genie is out of the bottle and somebody else will make it happen.
Really interesting times to be in IT.
Solaris 10 and Open Solaris have the concept of zones and containers. The computer runs a single Solaris instance but can run isolated process trees in zones which share common libraries but can be updated for dependencies independently. The containers concept (in conjunction with zones) allows a fair share scheduler to guarantee a service level for each allocated zone (CPU/memory sharing, etc). IMHO, must better than Virtuozzo, VMware and Xen.
Mainframe Edition, too?
Sounds like something Trigence already does. No need for OS-level virtualization in which you need to allocate tonnes of memory for an entire OS. Just allocate what the app needs. It encapsulates servers/services and the entire firesystem supporting it on both Windows and Linux. Their online demo is really, really well done. We looked at this product not too long ago because we were sick and tired of having our machines thrash under so many VMs that need X amount of resources (memory, disk space) just so they could run an encapsulated application requiring 10% of the resources.
I always wondered how you pronounced 'virtual datacenter OS.'
Now, I wonder if they'll ever announce this as a product.
Do not meddle in the affairs of sysadmins, for they are subtle, and quick to anger.
And, of course the obligatory photos of models pretending to be employees, happy customers, or drunken vagrants; who the fuck knows.
And why do the marketeers that they hire to advise them on their "onlin presence" insist on that shit?
Does anyone here get a boner when they see those fucking pictures of happy corporate people on every fucking corporate website?
I know Cisco has been trying to flog their VFrame (http://www.cisco.com/en/US/products/ps8463/index.html) technology which sounds very similar to this. Funny thing is, VFrame supports VMware itself, so im not sure how that relationship is going to continue.
Multithreading is not a big concern for server side developers, who are the ones who make use of datacenters this new "OS" claims to replace, nor are they under pressure to take advantage of all the CPU cores. In fact, most multithreading problems are taken care of by underlying application server platforms (apache, jboss, etc.). It's actually rarely a good idea to have multithreaded code handling single transactions on the server side.
That isn't to say that VMware's datacenter OS isn't interesting, but I hope they aren't trying to solve the multithreading "problem" like McAllister is saying.
I mean, go a talk to a CFO or a COO, and they'll [most of the time] politely complain about IT being expensive... IT becomes a true utility, with a pay-as-you-go menu that implements a coherent chargeback model that will bring a smile to the guys in dark suits.
Really interesting times to be in IT.
Or not.
Everything you've just written indicates that the chores which are currently being performed by 10 IT dudes might, in the near- to mid-term-future, be acccomplished by a single IT dude [who himself might be halfway around the world in Bangalore].
Can you say, "Buggy Whip Industry"?
PS: On the other hand, this might drive IT costs so low that it would make sense for the Bangaloreans to ditch their own physical infrastructures and rent virtual time out of the USA.
You can run all kinds of modern software on a mainframe. You get anything linux can run, but outside of linux you can use java, websphere, DB/2, IMS, messaging (MQ on a mainframe is amazing).
Yeah, you can use RPG, COBOL, PL/1 and that other crap. but why would you =-)
PHP is the solution of choice for relaying mysql errors to web users.
Linux can already do it.
Multiple Linux VMs run fine under z/OS, which of course is IBM kit. z/OS can already share multiple resources accross different physical data centers.
Oh, and word to the wise Linux system admins running under VM. Remove any stupid uneeded start up scripts. It doesn't make me a happy bastard operator from hell when I vary on the Linux partition and I have 60 servers thinking bootup is a good time to reindex their man pages.
When will VMWare work with 1394 devices and when will it be able to burn a DVD?