Mendel Rosenblum is one of the founders of VMware, chief scientist, and husband of VMware CEO Diane Greene. As such he’s got a lot of pull, getting his continuing vision of virtualization and systems management made reality.
He’ll be speaking in a few minutes out at VMworld 2007 in San Francisco. I’ll be liveblogging it in this post, so keep refreshing and I’ll keep posting.
8:30 AM – General session hall, Moscone Center halls B & C. Pop music playing, green and blue lights and graphics. As with the other liveblogging sessions a big thanks goes out to the VMware Community team for setting up tables, Ethernet connections, and power for us bloggers. Those guys are great and have really taken care of guys like myself. Thanks John, Robert, Brian, Anders, Cindy,
8:36 AM – A video of people at VMworld asking for virtualization in real life. “I’d like to snapshot conversations with my wife. That way when I say something stupid I can roll it back.” Hah! Awesome. Also talking about current wishes for VMware products.
8:39 AM – Karthis Rau, MC. Thanking folks for the party. Long tradition for VMworld to spend the last day on the future. Mendel Rosenblum on using mobility to transform your infrastructure, adding agility.
8:41 AM – Mendel Rosenblum. Going to talk about the state of virtualization, do some demos of stuff from VMware Engineering, and then we can predict for ourselves the future of virtualization.
Virtualization described as magic data center pixie dust. Not a helpful description. It’s more about complex machines interfacing with each other. You find an interface between them to interpose virtualization. As long as it still works like the old system you can get away with it. Adding levels of indirection, in CS terms. “All problems in CS can be solved by adding a level of indirection.”
(Funny, he’s nervous, got stuck, told a joke to break up the mental block. Awesome to see the human side of corporate leaders).
The first solution for the new layer of indirection was to cram more OSes on each physical piece of hardware. Second, because there is total control over where things are mapped we can do resource management and VMotion. Totally decouple them from the hardware to drive it as efficiently as possible.
Now we just adds pools of resources. The infrastructure makes the decision about how to use that pool most efficiently. Treat hardware as something you drop in.
8:50 AM – Rest of the industry really embracing this technology. An example: new hardware compatibility problems (making VMotion impossible to use). Causes partitions in the resource pool, makes life hard. Intel and AMD added capabilities to their CPUs to handle this, lie to the VM about the capabilities of the CPU so it remains backwards-compatible.
Talks about them asking for this compatibility early on. Asked the CPU guys to do this, they said “eliminate all the new features in the CPUs? What? No way!” And now we have it.
What sort of games can we play with this abstraction? Why can’t we move the storage around underneath the VM now? (my comment: SWEET, finally a public demo to confirm all those rumors out there). A demo of an ESX host with two data stores, unbalanced. Uses a script to connect to VirtualCenter and initiate storage VMotion. Graphs of the I/O to the data stores in real time showing the decline on one datastore and increase on the other. Relocation is complete. (applause!)
What does this get us? Leased storage is now easy to deal with. Capacity problems can be dealt with easily. Agnostic look at storage, you can move something from fibre channel SAN to a NAS device, no big deal. Excited about all these smart people working on placement problems for VMs. Now they can work on storage placement, too, optimizing that.
Next thing: virtual appliances. A review of the concept of virtual appliances, putting a stack of software together and building a “box” that you can download to do something specific, like CRM for your business. Gets rid of some of the traditional problems you see in an environment.
What about a marketplace for appliances? Appliances are really big, take a long time to download. What if you speed it up by adding “instant on” capabilities, so that you can stream a VM. Second demo, an engineer named Bruce showing the instant on appliances.
410 MB browser appliance, comparing the traditional download method to the streaming method. Starts a download on the traditional one, then goes to streaming one. “VMware OnDemand” the new product name, and a new stream manager built in. It’s prefetching what it needs, prioritize the blocks in the appliance that it’ll need. The appliance recorder analyzes the VM as you boot it and run some things, and then it orders the blocks appropriately.
Uses: marketplace, testing, ISVs, all sorts of things (my comment: ACE). Works with VDI, one of the big use cases.
9:03 AM – Software delivery models. Traditional (local only), SaaS (software as a service), hybrid (server-side management but local execution). See all of these models coexisting into the future.
Switches gears. High Availability. Describes a VM problem such that the infrastructure can notice the death of a VM and act on it, restart it. With the virtualization layer you can record the execution of the VM inexpensively, then take the log of execution and replay it to see where things went wrong. Present in Workstation now.
What if that is combined with HA to do Continuous Availability? Running a log of execution, sending it to a secondary VM to be replayed. (my comment: SWEET. Here’s my general DR solution. Plus, in a single stroke they killed a number of third-party products.). If there’s a problem with the primary VM or host the secondary can be activated seamlessly.
Demo from Dan from Engineering. An MS Exchange VM running LoadSim. Shows two VMs mirroring each other through the Continuous Availability feature. Mendel asks “If I pull the plug it’ll continue to run?” and goes looking around for the plug. Laughter. Plus the plug, the first VM dies, and in about five seconds the secondary takes over.
9:11 AM – Into the future. What are the hard IT problems to solve? Can take care of CPU & memory management now. New DR products now to manage site failures. New storage management tools. New hardware fault tolerance features. Can we guess what’s next?
Another thing to talk about is how the virtualization layer is responsible for the optimization of hardware. How do you do this in the most efficient way? In the future we can fine tune this process. We’re really conservative now, but as the algorithms get better we can do more. Power efficiency is a good area to tune for.
We’ve only scratched the surface with virtualization. Not much deployment, really, but as people see it the deployment tends to be rapid. Real bump up in efficiency. But it’s compatible with the way things are, so you can just slip it into your data center. Then you get opportunities to do new cool things.
9:16 AM – Thanks everybody. Karthik Rau again. Virtualization driving a complete refresh of the data center.
VMworld Europe 2008 is Feb 27-28 in Cannes, France. VMworld US 2008 is in Las Vegas. Powerpoint and audio for each session will be online.
Thanks all of us for coming. Safe travels.
(My comments: I won’t beat the pundits to the analysis but this is groundbreaking. More later).