Will Virtualization Undermine Network Equipment Vendors?

As you evaluate the promise of networking and virtualization stocks the question of how virtualization takes shape could make all the difference. The fortunes of companies like VMware (VMW), Cisco (CSCO), Microsoft (MSFT), HP (HPQ) and others can fluctuate as their offerings appeal to IT buyers.

And the first wave of virtualization projects in many enterprises have had very little to do with the network. That is not necessarily an encouraging sign. Recently I asked the question of whether or not most of today's networks could even keep up with the new automated systems.

That is why a recent blog discussion between Arista's Gourlay, Cisco’s Sultan and myself, followed by a blog from F5’s MacVittie is interesting. It is filled with implications for the future of the network (vendors and careers) and the evolution of IT.

The surprising blog conversation is likely an artifact of the success and shape of stage one virtualization (or virtualization-lite). With virtualization-lite system pros convert their legacy servers into images inside commodity blade servers (populated with virtual servers) and IT shops see an immediate economic benefit in the form of increased server utilization, responsiveness/flexibility and energy savings. That is pretty compelling, yet renders the network somewhat irrelevant.

In the short term continued growth in servers (now virtual machines [VMs] instead of “tin-wrapped software”) means larger blade servers with higher densities of virtual machines inside them. Servers can be created virtually and moved (within VLANs) in seconds, much faster than the days it could take with legacy servers and the network processes.

The business case for this first stage is compelling and the initial savings can be used to fund even more virtualization projects.

Imagine an operator at a 40s era switchboard trying to keep up with a moving cellular caller throughout the day? Imagine the cost of populating enough switchboards to manually track callers? That is the essence of the challenge that virtualization poses to the network today. More change, more complexity and more endpoints will drive more pain into a network that is manually managed.

Mouse click to create a new virtual server then wait two days to assign an IP address, etc. I don’t think so.

This breakdown between system automation and network manual labor is the core of what has made the network largely irrelevant to the virtual machine that has replaced the traditional server. VM economics can be so compelling that the savings from a single project can fuel additional projects and additional savings.

That is why network vendors will be rushing to develop the connectivity intelligence required to dynamically link applications, endpoints and networks. There are not enough bodies to throw at that problem.

MacVittie’s point in her blog (mentioned earlier) was that VM density was becoming the “new measure of IT efficiency”. If that trend continues the network becomes only tactically relevant to the data center, a development that would please some vendors and be troublesome to others.

Unleashing VMotion

However, there is more than a glimmer of hope in the eyes of networking vendors and professionals, and it’s called VMotion. It’s the ability of that virtual machine to easily move across VLANs (including data centers thousands of miles apart) in pursuit of potentially massive operational savings and scalability.

While the VLAN container represents tactical VM flexibility (spinning up and moving servers within a confined area), VMotion represents strategic flexibility or the ability to quickly pursue the lowest cost options for delivering services (electricity, taxation, labor, etc.).

Rather than building a very expensive class A data center, cloud builders can build a global mesh of smaller ones and see both capex and opex (ongoing) savings. Local changes in rates or regulatory events could be addressed in minutes instead of years. That’s a pretty powerful promise.

As Google (GOOG) and Amazon (AMZN) and other cloud providers build massive facilities (typically in areas where electricity and real estate is inexpensive), enterprises could be very cost competitive by building meshes of smaller and less costly facilities continuously optimized for cost based on a global spread of expense options.

While networking vendors have been on the sidelines for the spread of virtualization-lite, they will be front and center as VMotion is unleashed. That level of dynamic movement will require new investments in automation and management, especially around addressing.

Like during the emergence of VoIP, the role of the network has been misunderstood (as blogged recently). Large scale VoIP projects grabbed the headlines as enterprises ultimately discovered the network impacts and made additional network investments. Virtualization will be no different as these blade servers continue to be populated with more VMs.

When the network vendors solve the VMotion challenge they will create a new dynamic in virtualization that could impact the fortunes of both virtualization and networking players. That work may be already underway, as a team of “network industry legends” convenes in early September at SRI under the title: “Pipes and Clouds: The need for Infrastructure 2.0 (The Intercloud Challenge)”.

Networking vendors are also stepping up. Earlier this year Cisco announced their Unified Computing System with advanced networking capabilities. Startup Arista Networks joined the infrastructure 2.0 race recently with an impressive vEOS announcement:

"With vEOS, Arista and VMware customers will now be able to more easily manage their growing converged networking infrastructure and enable easier mobility of virtual machines," said Jayshree Ullal, president and CEO of Arista Networks. "Arista’s vEOS is fully compatible and capable of co-existing with the VMware vSphere™ 4 vNetwork Distributed Switch while providing consistent administrative experience to both the network and virtualization operator."

Clearly the network industry is waking up to the issue and figuring out how to take virtualization to the next level. At Infoblox we were already planning a webinar with Nemertes, Cisco and VMware executives for late September on virtualization and the future of the network, so the contrast between the blogs and these network industry announcements has set us up for an interesting session.

Disclosure: Long VMW

Comments

[Joseph L. S...:]

Brave New World...

Aug 30 12:25 PM

[ManAboutDallas:]

The first virtualisation product is still the best: IBM's z/VM

Decades ahead of its time when introduced in 1972 - YES!, 1972 - z/VM - or VM/370 as it was called then - for years was a solution looking for problems.

Now, the problems are beating down its door. Which would you rather manage: a warehouse full of blade servers or one - yes, ONE! - new "10"-class mainframe running z/VM hosting 10,000 virtual Linux images all plugged into your "cloud".

Simple choice, really.

Aug 30 04:39 PM

[Alex Filonov:]

Why wait 2 days for IP address when you can change DNS entry in less then an hour with full propagation?

Aug 30 04:48 PM

[LaughingTrader:]

The growth of virtualization seems to me to have a fairly predictable impact on the various players in the tech market. First, the computer hardware vendors (Dell, HP, etc) are hurt by this trend as a single box can be configured to "look" like multiple machines on the network. The network vendors are helped by virtualization as the number of nodes multiply with ease. Lastly, software vendors are helped by virtualization, again because the number of machines (licenses) is easily and inexpensively expanded.

Long term, I also think virtualization hurts Microsoft, even though in the short term it helps them. This is because I forsee a time when the operating system truly does not matter. The OS will be distributed bundled with the applications as a virtual machine that users will just run alongside other virtual machine/application bundles. Users would routinely be running OS/application bundles featuring different OSes, side-by-side, without even being aware, or caring, that one application is running in Linux while another is running in Windows. As this plays out, Microsoft will lose the massive leverage its gains from Windows' dominance.

Aug 31 08:57 AM

[MikeOz:]

On Aug 30 04:39 PM ManAboutDallas wrote:

> The first virtualisation product is still the best: IBM's z/VM
>
very interesting comment - but doesn't that beg another question ?

Which is that a warehouse full of commodity servers cannot be the most efficient way of providing the cloud service, even if it were 10,000 virtual linux boxes on a single z/VM.

As I understand it, google, amazon etc are stuck with their current solution because their software cannot be ported to a mainframe environment. But that might change ...

And in any event I have heard Google managers say that they see no place for virtualisation in their operation - i think they already switch processing around the world to suit with their load distributing systems.

Aug 31 10:42 AM

[ManAboutDallas:]

That's just it, Mr. Oz....just the point... a single mainframe running z/VM is the host of those 10,000 Linux images, NOT a warehouse full of blade servers. I'm not sure where or what your "begged question" really is. To put that in better perspective, the new z10EC "Enterprise Class" zSeries server is about the size of a Sub-Zero refrigerator:
www-03.ibm.com/systems...


On Aug 31 10:42 AM MikeOz wrote:

>
> On Aug 30 04:39 PM ManAboutDallas wrote:

Aug 31 11:37 AM

[MTimJones:]

Google's example isn't quite conducive to virtualization like other problems. Google's problem is compute and I/O intensive (map/reduce computation). The problems associated with virtualization are those that tend to be idle much of the time, and therefore can share a physical server. So, agreed, but Google's problem is fairly focused (to the point they define their own servers).

On Aug 31 10:42 AM MikeOz wrote:

>
> On Aug 30 04:39 PM ManAboutDallas wrote:

Aug 31 11:43 AM

[Zmartmoney:]

I see a lot of comments here that are seriously lacking in technical understanding of the factors involved. My advice - don't use any of the advice given here for a real investment decision. The battle of the titans is coming soon and it will be between Cisco, HP, and Dell. IBM has been backing out of the fray for more than a year. The philosophy of approach will be decided - not quickly and not without a lot of pain for everyone involved. There will be clear winners, but not anytime soon. It's also a complete unknown whose philosophy will prevail. Good luck...

Aug 31 12:31 PM

[Gregman2:]

So, connectivity intelligence=infoblox. Anything else?;-)

Aug 31 12:49 PM

[Gregory Ness:]

Z:

There are a few more titans in the battle but I think you've identified those who could be called "strategic" (value add) versus "tactical" (cheap IT service). I think how IT evolves and how well the network embraces the virtual infrastructure may have a lot to do with who wins the most.

Interesting blog by Chris Hoff: www.rationalsurvivabil...

Thx
G

Aug 31 01:01 PM

[Gregory Ness:]

Greg:

More than Infoblox. Addressing is a core issue... but work needs to be done with multiple players. Ultimately the network, systems and endpoints need to know each other's status on a continual basis. That's bigger than the automation of DNS/DHCP and IPAM etc as you likely know.

Thats why I think the SRI event is happening and Infoblox is a core part of the conversation. It will take a collaborative effort among multiple companies to establish connectivity intelligence in a way to build infrastructure 2.0.

This matters (the shape of this evolution) because it will impact who is relevant and strategic to virtual and hybrid infrastructure (the future of networking as we know it).

Thanks for the shameless plug opp... but I'll pass and answer that Infoblox is very relevant/strategic/nec... to this direction but not the entire solution. That is why there are several companies talking about this issue and SRI is sponsoring an event.

Stay tuned. :)
G


On Aug 31 12:49 PM Gregman2 wrote:

> So, connectivity intelligence=infoblox. Anything else?;-)

Aug 31 02:13 PM

[Gregman2:]

"It will take a collaborative effort among multiple companies to establish connectivity intelligence in a way to build infrastructure 2.0."
You nailed it. By the way, excellent article.

Aug 31 02:23 PM

[Gregory Ness:]

Thanks. I'm tempted to post the agenda for the Infrastructure 2.0 blackboard session at infra20.com to "address" some of htese comments. Let me check with the event sposnors/champions.

Thx
Greg

Aug 31 02:34 PM

[Wisdom vs. ...:]

the cloud and virtualization may actually reduce CSCO/DELL/HPQ's total rev, so the best investment may turn out to be short the sector; and, VoIP implications: more powerful, cheaper systems are now available, reducing total rev additionally

IMO what most biz really want is a pipe and a bill; the companies that can offer that first will win whatever is left to win. CSCO service revs and consulting firms will be hurt, and IT professionals will get laid off or moved to industrial cloud facilities; however, this transition will take many years.

two big blockages are transport and security. 100Gb transport should dominate in two years, and security/encryption MUST improve for the cloud to work, so those are two areas to place your money.

Aug 31 03:23 PM

[Gregory Ness:]

Wisdom:

That is certainly one of the schools of thought... driven in part by the success of the VLAN. The second school shows a peak to the tactical VLAN payoff and a disruptive evolution of the network.

VoIP ultimately drove more network revenue because it increased hte demands on the gear. Very few pros considered those impacts for a variety of reasons... but then they happened.

Thanks!
Greg

Aug 31 03:35 PM

[Phil Dewey:]

Comparing VOIP's impact on the network to Virtualization is not valid. VOIP put new, real time stresses on the network. Virtualization removes them from the LAN and puts them potentially on the WAN. The idea that there will be a valid use case for moving a VM across a wide distance is goofy (much less a running VM). Data will move the, VMs will stay where they are. Taking advantage of cheaper compute resources elsewhere will mean getting a COPY of a VM to materialize at the lowest cost facility and the data there to use it.
One won't move a VM but rather a description of it, a formula of how to construct it and the data it needs to do it's job.

Aug 31 08:25 PM

[Gregory Ness:]

Phil:

I wasn't saying that the comparison was based on an identical set of impacts, but rather that in both cases the demands on the network were underestimated. If you are in the virtualization-lite camp then you perhaps won't see material demands, except perhaps related to security and vmotion (I/O) within the vLAN. Some firewalls and ids solutions might require hairpins and extra processor cycles depending upon how they're deployed. With VoIP the network challenges were discussed openly in the press... Dow, City of Houston.

Thanks for the comment
Greg


On Aug 31 08:25 PM Phil Dewey wrote:

> Comparing VOIP's impact on the network to Virtualization is not valid.
> VOIP put new, real time stresses on the network. Virtualization removes
> them from the LAN and puts them potentially on the WAN. The idea
> that there will be a valid use case for moving a VM across a wide
> distance is goofy (much less a running VM). Data will move the, VMs
> will stay where they are. Taking advantage of cheaper compute resources
> elsewhere will mean getting a COPY of a VM to materialize at the
> lowest cost facility and the data there to use it.
> One won't move a VM but rather a description of it, a formula of
> how to construct it and the data it needs to do it's job.

Aug 31 10:32 PM

[itosolutions:]

You can download Vsphere for 60 days at ITO Solutions...

itosolutions.net

Sep 02 01:36 AM

[LeitM:]

VMotion is a specific vendor's offerings (granted that the vendor is the market leader). Xen and Microsoft offer similar capabilities now - so it might have been more vendor neutral for the article to say "Live Migration" instead of "VMotion".

Sep 02 06:27 PM

[Gregory Ness:]

Leit:

That's certainly a fair comment. Although I do think that VMware has done more with the network vendors than hte others in terms of grappling with the network effects of virtualization. Point well taken.

Sincerely,
Greg

Sep 02 07:20 PM