The Real Impact of vSphere

Another milestone day today -- VMware announces their "cloud operating system" -- vSphere.

Much will be written about what it does and how it works.

I thought I'd step back and talk about what it means -- at least, what it means to me.

It's Not Just Cloudy Marketing

I'm sure that there will be some who will view VMware's choice of cloud-ish monikers as just another example of buzzword marketing.

I'd encourage these people to look a bit deeper, and hopefully learn to appreciate a few simple axioms:

First, it's pretty easy to make the case that the only way that enterprises are going to leverage internal and/or external clouds in an evolutionary manner is by virtualizing their existing workloads. 

And with vSphere, VMware is now poised to containerize 99% of the workloads that are out there.

Second, you're never going to build an internal cloud by simply standing up individual hypervisors.  You're going to need multiple hypervisors working together to achieve load balancing, QoS, high availability, etc. 

With vSphere, the differentiation here becomes even more pronounced.

Third, creating the appearance of a cloud to your end-users (whether internal or external), has a lot to do with self-service provisioning and expedited workflows. 

We see that "cloud-style provisioning" with this version of vCenter.

Fourth, no one is going to trust any cloud -- internal or external -- unless they get their 5 or 6 "nines" availability.  In addition to VMware HA, DRS and SRM -- we've now got a new flavor of software fault tolerance to drool over. 

It's not hard to argue that the vSphere-based cloud will potentially be more highly available than traditional computing approaches with this release.

There's more, but I think the picture is now clear -- all of the technology-based arguments for not virtualizing the majority of your workloads are now off the table.

It's now up to the "3 P's" (people, priorities and politics) just how fast this is going to happen.

The Ball Is Now In IT's Court

I see the rationale for staying with legacy UNIX environments eroding very quickly. 

One of the old arguments was that RISC processors provided a performance advantage, and you needed a customized version of UNIX to support the power of the architecture.  I, for one, am not betting against Intel's Nehalem and future offerings.

Another argument was that you needed big UNIX to handle really big workloads.  Novell's SUSE running under vSphere now makes a very credible alternative to legacy UNIX approaches for running very large workloads indeed. 

And, let's be honest, how many of your apps need more than 8 virtual CPUs, 256GB of RAM and 200K+ IOPS?

VMware For Tier 1 Apps

As long as we're talking about impacts, I think there's a certain school of IT thinking that's due for a refresh. 

They probably never fully realized that -- before long -- VMware was going to be able to handle big, important workloads.

And as a result, they built virtualization infrastructure "on the cheap" -- using second-class servers, networks and storage -- not to mention thinking through issues like backup, security, performance management, disaster recover, service delivery management -- all the stuff that gets really important in a Tier 1 environment.

Sure, doing things cost-effectively is great, but only if you can get the job done.  And that's exactly where there are starting to be a few problems.

I've already met a few customers who've realized that their previous infrastructure decisions won't let them go to the next level of virtualizing the majority of their workloads.  For a few, it's rip and replace time. 

That's not good -- and can be avoided.

The Bottom Line

I hope most IT thinkers will look beyond the marketing noise, and realize that virtualization has taken a major step forward with the vSphere release.  Virtualization has moved far beyond a neat way to consolidate servers, and morphed into a new way to build data centers and consume IT as a service.

Whether you choose to call this cloud (or not) the outcome is the same.

Once again, virtualization changes everything.

Comments

[nate]

(Vmware user since pre 1.0 back in the late '90s)

Any idea if vSphere 4 fully supports round robin load balancing for storage? I have read hints that it would though I read through a bunch of stuff this morning on it but didn't see anything specifically called out for it. The two things I am looking for is this and ESXi boot from SAN(working on deploying a new Dell blade system for ESX/ESXi right now). It appears ESXi boot from SAN is in vSphere 4 it is called out in their features document.

Also fast SAN failover, talking sub 5 seconds here. I don't want my entire VM environment to freeze for a minute or more if a switch or a controller decides to die(or if e want to do a software update). With today's active-active controllers you should be able to fail over very fast(this works fine native in linux with device mapper on FC or iSCSI).

Until active-active multipathing and very fast storage failover vmware probably won't be the best for the
highest availability apps.

Our current vmware infrastructure is pretty small still, 3 servers doing the bulk of the work, the 4th is mostly idle (total 32 cores/ 220GB ram), CPU usage can kick up into the 25+Ghz range, memory usage around 120GB typically. Using Vmware Foundation.

I have 3 vmware data volumes, though I/O wise all 3 combined volumes average about 100 IOPS and about 1.8MBytes/second of data transfer(periodic spikes here and there). There are a few databases in the VMs but they are all using RDM, not VMFS volumes. About 55 VMs in total, about 310GB of data written(3x1 TB of thin provisioned volumes total), a good chunk of the I/O is from/to the NAS cluster(which shares the same spindles).

I suspect this new blade system(128 cores, 512GB memory) will drive that space usage by another couple TB over the near-mid term, most of this will be the basic ESXi, with a few systems running Foundation, everything fiber attached.

[Chuck Hollis]

Nate

You really, really want PowerPath/VE. See latest post. Most definitely worth looking at.

-- Chuck

[aenagy]

"... with vSphere, VMware is now poised to containerize 99% of the workloads that are out there"

Unfortunately the real barrier will not be technical, but political. Support from those critical nitch application vendors who just will not support thier app on any virtualization solution even though they develop on that same virtualization solution. Customers are effectly being forced to assume that they should not have to. Win over the the nitch software vendors to get them onboard then we'll see about that 99%.

[Chuck Hollis]

Aenagy:

Very valid point, so thanks. From a technology perspective, there are very few -- if any -- barriers left for the vast majority of IT workloads.

The software vendors have to come on board. There's also serious resistance from certain business users that demand "their server" -- I hear stories about this continuously.

I guess my point would be -- the technology is there. Now it's up to all of us to work through these barriers, and realize the potential of what's now possible.

Thanks for writing!

-- Chuck