The Cloud Will Force Networking Vendors to Change Their Stripes

21 Comments

When a company builds a web site in the real world, they assemble servers, routers, switches, load balancers and firewalls, wire them up, configure them and go live. But when that application moves into a cloud environment, things change. In a cloud model, the customer isn’t dealing with physical equipment. So who handles all the wiring? And more importantly, how do networking vendors get paid?

Many operational clouds still require their customers to corral their own machines, however virtual. Amazon Web Services is a good example of this. To build an application, the operator still needs to do what they do in the real world — assemble servers, routers and switches to make a data center — only this time, they’re configuring virtual servers instead of real ones.

On the other hand, development clouds like Salesforce.com or Google’s App Engine hide the underlying machines, and handle all the networking equipment — virtual and real — on behalf of their customers.

Either model means a big transition for the makers of traditional networking equipment.

Option 1: Virtual appliances

In a cloud world, the routers, firewalls, and load balancers run inside “virtual appliances” — virtual machines pre-configured to route, block or distribute traffic. Cloud users still have to configure and provision them.

Open-source software dominates the virtual appliance world. For load balancing, Pound is one open-source alternative. For firewalling, there’s IPChains; for routing, Xorp. Some clouds already include these components: Cloud builder 3Tera, for example, offers users a catalog of data center components, including many open-source elements, in its default configurations.

Some vendors stand to gain from a move towards virtual appliances. If you want the kind of service and support you’d get from a vendor, Vyatta does for networking what Red Hat did for servers and MySQL did for databases. And while Checkpoint makes equipment, its software-based firewalls are more easily deployed in a virtual environment than many of its appliance-only competitors. The pendulum swings back to software.

If equipment vendors want to target this market, they need to convert their equipment and licensing models to virtual appliances and differentiate themselves based on software functionality rather than on box color or port density. Companies like rPath and jumpbox both specialize in turning traditional software into virtual appliances.

Option 2: Sell to the cloud operator

But what if the cloud handles the network equipment? This is the case if you’re using a development cloud like Salesforce.com or Google’s App Engine, or if you rely on a turnkey cloud like Joyent or Heroku. The networking equipment vendor sells to the cloud operator.

Which is No Fun At All.

Selling to a utility is notoriously challenging. Carrier sales cycles take months or even years, during which margins get squeezed razor-thin. At the same time, the list of requirements grows dramatically. Because clouds buy tremendous amounts of equipment, they have strong negotiating power. And they often build their own management tools, removing the differentiation a vendor’s software provides.

To make matters worse, clouds may need different equipment. Vendors are innovating, of course: Cisco’s new high-end switching platform, the Nexus 7000, seems well suited to this task. Further, the company has had strong carrier sales since its acquisition of Stratacom in 1996.

Some clouds may even find they have the expertise and economies of scale to build their own equipment. By buying directly from chipset manufacturers and using open-source libraries, they can bypass equipment manufacturers entirely.

One way or another, it won’t happen overnight. While the advent of utility computing is sure to change the networking industry, it will be some time before the trend puts a dent in enterprise IT equipment revenues. Less than 2 percent of CIOs surveyed by Goldman Sachs considered cloud computing a priority.

But someday soon, that load balancer you deploy may be a virtual one. That means two big changes for equipment vendors. One, selling licenses instead of boxes; and two, repositioning their sales forces to sell to telcos and utilities.

21 Comments

Lori MacVittie

Nice post, Alistair.

It’s interesting that you mention Cisco, but don’t mention that cloud computing providers like Joyent are already taking of advantage of F5’s solutions to build a strong cloud computing infrastructure (Joyent’s Technical Specifications calls out F5’s BIG-IP).

Flexibility, scalability, and operational efficiency of both the infrastructure and the applications that it will deliver are paramount for helping cloud computing providers achieve their goals, which is why folks like Joyent are relying on F5 to build out their cloud computing infrastructure.

Cheers,
Lori

Brian Thompson

Great writeup, I eagerly await to see Cisco’s response to the cloud…will they come up with an original plan to compete in that space or will they buy an original plan from another company? Their lust for acquisitions has given me headaches over the past few years.

The other question is, who will the big performance analytics players be in the cloud? How do you “blame the cloud” if you have only a virtual (limited) view of its infrastructure?

Michael Sheehan

I thought I would jump in here as well as the Technology Evangelist for GoGrid. Data centers in the cloud presents plenty of new configuration requirements for the IT professional. When we developed GoGrid (Cloud Infrastructure provider), we had over 7 years of traditional hosting experience under our belts. This is part of the reason we made sure that we launched with more than simple cloud servers alone, but also included some virtualized appliances/hardware as well. An end-user is looking for a 1 stop shop when it comes to cloud computing. We developed a way to programmatically hook into f5 load-balancers and provide that as our free load balancing option for our cloud environment. The framework within GoGrid is set to easily provide other virtualized hardware or appliances. But LB is just an example to your points.

My view is that you should really be able to use whatever option you want, whether it be a solution by a single cloud provider or by chaining other providers’ offerings together (hardware, computing, storage, or the like).

I’m sure the networking professionals and vendors will be earning their pay as Cloud Computing makes us burn through IP addresses exponentially and all vendors will be looking to figure out the next y2k with IPs. IPv6 will help for a long while but then what? Another subject altogether.

I enjoyed the read. Thanks!

-Michael

friarminor

And the cloud future just as it seems to close some doors also opens a whole new expanse of businesses. Judging that there wouldn’t be a sort of ‘standard’ on how the ‘ideal’ cloud will operate, expect a lot more entrants to tech fields both in hardware and cloud management tools.

Same old tech adage: Innovate or be left behind.

Thanks, Sinclair!

Best.
alain
http://www.mor.ph

Dave Roberts

In full disclosure, I’m the VP of marketing at Vyatta, so I couldn’t agree more.

At Vyatta, we see many of our customers wrestling with the issues surrounding these new deployment models. Many of them are engaging with Vyatta precisely because we aren’t bound to the old an-appliance-is-a-physical-box model and can accommodate the new requirements easily.

That said, the folks that say that cloud computing is overhyped are also right. While there is no doubt that cloud computing will gain over time, this is not going to be a 1-year transition of the market. This is a long-term trend and even then there are going to be plenty of reasons to “go physical, not virtual” for a long time.

At least a Vyatta, the thing we’re most excited about is the opportunity to address both models. Customers can choose which is preferable for them, sometimes on an application-by-application basis, and Vyatta can deliver a solution that with consistent features across multiple deployment models.

Bert Armijo

Well thought out post, as usual.

Having spent a decade building networking gear before starting 3tera, IMHO networking vendors face more than an economic shift in selling to cloud operators. Cloud computing will have very different requirements than traditional customers. Cloud is about scale and homogeneous resources, while traditional data center purchasing decisions have been based more on features. Thus cloud computing favors large, high performance, switches with relatively few features. Force 10 has been at the forefront of pushing both port density and performance, so Joyent’s choice of Force 10 over their far larger competitors exemplifies exactly what you’re writing about.

Douglas Gourlay

Agree with your sentiment that things will have to go virtual- we recognized this a good while back delivering virtual contexts on firewalls in 2003, virtual load balancers in 2006, and then as you indicate the Nexus 7000/5000 this past year.

Virtualization changed the velocity the servers could be rolled out, moving provisioning from a physical task to a configuration task. As such the speed of the change was cut significantly, ‘months to minutes’ was one slogan I remember.

Clouds will be the evolution of this virtualization- virtualizing not only the devices but their location, and potentially even the presentation layer so I can achieve the Nirvana-esque goal of Any Content/Application on Any Device, Any Where.

There are some types of devices and functions that don’t virtualize very well yet- those being ones that require something ‘special’ in hardware. Like SSL Offload processors or IPSEC VPN termination, Deep Packet Inspection, etc. x86s are always getting faster and sure we can throw cycles at these workloads but if specific processor types perform the job more efficiently because they are purpose-built for it then I would go with the efficient choice.

dg

Experienced Realist

If HP is involved in this venture, it is certain to miss expectations. They should stick to their knitting of selling servers and printers. Point out one web 2.0 project that HP has helped launch (other than just providing boxes)…

Alistair Croll

@Gred: RIA and client-side computing is definitely something to watch. The whole area of synchronization is at the core of most big data problems these days, and may reduce demand for traffic if bandwidth pricing changes.

@BR: I tend to agree that Netscreen and others ate Checkpoint’s lunch when they failed to adapt their pricing strategy, which ultimately led them to license many of their products to others. But now that the software/hardware pendulum is swinging back (in on-demand platforms, at least,) that slowness may be an advantage for companies that resisted the move towards equipment.

@Jason: You’d know better than I. I’m lumping “pure” clouds like yours and “cloud on cloud” environments like Heroku here because in neither case does the cloud buyer have to provision, say, a load-balancer. And while folks like Cisco have definitely honed their carrier sales skills, I think we’ll see less sale of specialized networking equipment to enterprises in the coming years, particularly in the small- and mid-tier markets.

All of the big networking companies have midrange products, but IMHO we’ll see Cisco selling Linksys at the low end, and a Nexus at the high end, and less in the middle.

As for Force10, that’s sort of my point — Joyent can make a technology, rather than marketshare or support, decision because it’s in the business of operations. Most enterprises go with what’s safe/known/predictable. So an incumbent vendor has more to compete against when selling into a cloud. I would argue that most enterprises aren’t evaluating their networking equipment based on “ease of programmability.”

@Aaron: I think the clouds will buy programmability (as Jason of Joyent pointed out above,) which is generally a higher-end feature (your Netgear isn’t that tweakable ;-))

But in general, I think the future of new entrants in cloud computing is specialization — clouds just for HIPPA, or PCI; clouds just for a certain geography or legislation; clouds just for specific app stacks (like Engineyard for Rails); Clouds with a specific pricing model (per-transaction, per-month, per-seat); and clouds with APIs to back-end verticals (travel, auction, media delivery, etc.) Probably something for another post.

Joe

I’m sick of giga’s evangelistic preaching about the cloud. the cloud is convenient, but there are way too many bottlenecks, security issues to make it realistic for many. 100 years from now I’m sure people will be still be using some sort of machine to store data locally, especially with the rising affordability of terabytes of space. Sure there is google email and docs, but that’s about my limit, and I back that up regularly. There are way too many companies out there boasting to store your data, passwords, it makes me sick.

The simplicity of owning a 16-32GB usb stick to move files around cannot be ignored.

Aaron deMello

Great post, Alistair. One thing that works against the large utility cloud vendors is that their need for homogeneous environments to keep maintenance costs low and to increase buying power means its likely they they won’t be buying the latest greatest technology – which may mean that the clouds themselves won’t feature bleeding-edge functionality. That in turn might create an opportunity for “Cloud 2.0” companies that compete with Amazon et al not on price, but on core capabilities.

Jason A. Hoffman

I’ve yet to see “Virtual appliances” around networking that really have both ease of use and great failover. Or a proper matching of that software with the hardware that it sits on. Going with open source libraries and software in these scenarios (on your own) means that you’d better be ready to actually do nearly everything … on your own.

Networking vendors have been selling into telcos for quite a long time, selling into any other “utility” is nothing new. I don’t think they’re terribly worried about it.

I don’t believe Heroku buys any hardware, they’re cluster shared rails hosting with a development GUI on EC2.

As someone that does buy all of our own networking gear, Joyent uses Force 10 and that’s actually a technology choice, and not drive by the price of the gear. The cost of networking gear (within normal ranges) doesn’t impact the cloud model much. So our networking choices are driven by technology and the needs for resiliency, wire speed, 10gig, ease of programmability, quality and service.

Now if 40% of the world’s applications were running on a single cloud, then there would be implications for everyone not just companies making networking gear.

BR

“…while Checkpoint makes equipment, its software-based firewalls are more easily deployed in a virtual environment than many of its appliance-only competitors.”

Au contraire. If there is any poster child for companies who’ve failed to adapt to new ways of delivering products, it’s Check Point. If they had their act together a decade ago, the “appliance-only competitors” would never have existed. I don’t see Check Point being on the cutting edge of selling to cloud vendors, if and when the opportunity arises.

Greg

If metered bandwidth happens the “cloud” might die a quick death. If the ISP’s get their way, which is likely, then technologies like Google Gears and Yahoo’s BrowserPlus will be hot as they allow you to store data on the client machine and use the computing power of the client machine rather than sending all that data back and forth.

Comments are closed.