Quote:
Originally Posted by NikFu S.
Hmm, I don't see how anything is possible with "virtual infrastructure" that is not currently possible without it. Virtual machines and emulation are hardly a new idea, and even with virtual machinization it's not going to increase the ability of the hardware to multi-task.
Once the other companies release their own versions of the hypervisor, VMware's stock is going to plummet. I don't see any use for this at all outside of large server-side businesses and if anything it will bog down a system that uses more than 30% of it's resource at any given time.
You can see on their own site, on their own benchmarks, increased performance hits it's peak early, and begins to fall almost as rapidly not unlike multicore processors.
http://www.vmware.com/overview/perfo...enchmarks.html
I suppose if you run a VM you can run a secure "whatever" where ever fairly risk and worry free, but it seems pointless to me. |
Disclaimer: I do VMWare for a living and am a VCP (VMware Certified Professional). As a result, I DO have a vested interest in the success of the company and product... but I'm also Citrix Certified and working with Xen a lot... so take that for what it's worth, too...
Contrary to your comment, virtualization DOES increase the ability for a given physical box to multitask. Particularly when dealing with Windows systems, the ability for the OS to arbitrate tasks falls apart as you scale up (number of applications). That's why most people actually end up scaling out; that is adding additional systems instead of having multiple processors and so forth. Typically, each server in an environment only runs one or two major tasks, with the rest of the system committed to running the rest (authentication, disk subsystems and so forth).
Now, as you scale out like this, the problem becomes one of load. A typical Windows server is
less than 8% utilized (and this is from my own studies, not VMWare's). This means that for an average server-based application you're wasting 92% of the server's computing power. Note here that I take computing power to be a combination of all factors; memory, CPU and I/O utilization... not just CPU utilization. For a company paying $10,000 for a server, that's a loss of $9,200 per server in the datacenter.
VMWare allows you to stack 6 or 7 of these systems on top of one another, leaving some overhead for growth in one physical box. It maintains all the advantages of outward scaling (application isolation, multitasking and so forth) while getting rid of the down side (wasted computing cycles).
Now the reality is obviously a LOT more complex than this. If it were this simple then we wouldn't need people like me. However, as a rough guide this is good enough.
I build systems that run in an high-availability backbone, each physical node running (normally) 15-20 guest machines... mostly Windows but some Linux and at least one Solaris box that I know of. So, across our current farm with 160 Virtual Guests, we have nominally 10 big beefy servers instead of 160 2U boxes in our racks (our standard stand-alone server used to be a DL385).
Now, we actually run our VMWare on highly powered IBM blades connected to a SAN. In fact, with technologies like VMotion (look it up!) we can actually take down a server for patching or hardware replacement without taking down ANY of the hosts on the server itself. Put it in maintenance mode, move them to other nodes and we've just moved a client-accessed server to another physical server
without downtime.
I'm sure you can see the value in this to corporations. It's huge, and I paid for my own VCP exam because I believe in the future of virtualization. Maybe I sound like a zealot... but it REALLY is cool tech.
MSG me if you want ot discuss further...