Enterprise Technology By Phil Edholm

Guest Blog - Will the Computer be the Network?

When we started the Nortel Enterprise Blog, one of the goals was to develop a forum for ideas from the larger Nortel Enterprise Community. Now that the blog is running and achieving some level of following, I thought it was time to introduce the first guest posting.

This post comes from Brad Black. Brad is Leader Security Solutions Engineering in our Enterprise Solutions Engineering group. Brad is a CISPP certified security professional and focuses on a variety of security issues from VoIP to data centers. He also has extensive experience in architecting campus, WAN, and data center solutions for Nortel customers.

From Brad:

Many years ago, one of Sun Microsystems' original employees, John Gage coined the phrase, "The Network is the Computer". This served as a powerful vision for Sun throughout the 90's and beyond. It always struck me as not quite right - Isn't the network the network and the computer the computer? Each had a specific job and when combined in the right way, over IP/Ethernet networks of course, computers and networking became an integral part of business, personal and cultural life.

A key part of what Sun was emphasizing was how you could distribute application components throughout the network. This powerful concept has evolved over the years into web services and service-oriented architecture.

Today, as I talk to customers about their vision of application delivery, a key trend with most larger enterprises is data center consolidation and virtualization. Some companies are further along than others, but the benefits of machine virtualization in terms of reducing wasted resources are very compelling. With many application servers running at just 10-15% average utilization, buying, deploying, managing, powering and cooling hundreds of servers just doesn't make sense. And when you move to a virtual machine environment, provisioning new virtual servers, backup/restore and disaster recovery get a lot easier too.

In some environments we see tens of virtual machines running on a single physical server. And this is not just about server virtualization. Client virtual computing is also taking hold such that the data center houses both application server VMs and end-user workstation VMs accessed via remote desktop protocols. With client virtual computing, you gain a number of security and application management benefits and decouple the end-user device from the IT infrastructure.

I can see the point in the near future where the entire data center can collapse into a pair of high end servers. You would really only need one except for availability considerations. Think about it - multiple application tiers (web front ends, application servers and databases) and client VMs all running together on one super-server. In this scenario - the network too moves "inside" the server providing VLAN isolation, routing, stateful firewalling, intrusion detection and even server load-balancing for VMs. The networking and security functions are not optional here - think about how to detect and contain a compromised VM hosting an attack on another VM on the same server.

When this model is taken to an extreme the computer is the network. Backed-off slightly, the computer is the data center. But I doubt that we'll see such a complete implementation of this vision. In the case of server virtualization, we still need another complete data center that is geographically separated to reduce the risk associated with fires, floods and other localized events. And we need a "real" network to connect those sites. Beyond that, network latency can still stand in the way of a true "global" data center for distributed international enterprises. In the case of client computing, virtualization is applicable to some information and knowledge worker applications but not others requiring mobility and off-network use. Phil needs to hammer out blog entries at 30,000 feet and the package tracking application needs to reach out to the delivery truck. And even the thinnest of clients will need a network to reach the data center.

In the end, hyper-connectivity is making the world a smaller place, connecting more people, applications and even things together and networks are getting bigger, not smaller. Even as machine virtualization takes hold and presents new challenges for those building networks and applications I still would say, "the network is the network".

Comments

  1. Phil & Brad - you’re approaching this the wrong way, IMHO. The question you’re asking is moot and obsolete. The real question needs to be application oriented. For example, “how do I provide ubiquitous and seamless access for application X?” Figuring out the answer to that question will factor in all the factors that will answer your obsolete question including cost, architecture, access, etc…. And when you combine the answers for the top applications, you’ll find trends that will enable infrastructure efficiency. But if you continue to try to develop a network/computer strategy independent of applications, you’ll fail. Todays world wants apps, not architectures. Provide something effective and quickly or get out of the way.

  2. Yep, the network is still the network (the sum of the hosts, servers, applications, transport and protocols).

    I disagree that everything is going to collapse onto a pair of geographically diverse servers (except perhaps in the smallest of networks). Even then, I expect so-call “cloud/grid computing” running across many many servers to be more cost effective and survivable than “per network” data centers.

    Virtualization will surely be a part of it, but the VM is nothing new and dates back to the IBM3270 days. We still haven’t seen the network collapse back to mainframes :)

  3. Nortel Watcher; I wish I had said what you said.

  4. Nortel Watcher: I completely agree that it is “all about the apps” and it is clear that any IT strategy must consider those apps and the business drivers. But ubiquitous and seamless access to those apps is not enough. Every week, another Enterprise customer is telling me that they are using virtualization to consolidate and optimize resources. So an absolutely key part of the equation is how can you build and run an IT infrastructure and do it in such as way that minimizes total cost of ownership and maximizes value and agility? To answer that question - you must consider the network/computer architecture in the context of those apps. So the question is far from obsolete when you get down to the business of planning your data center strategy and vision.

    Many: Thanks for the great definition of “network” - this is exactly how end-users think about it which is why when something breaks, they say “the network is down!”. :-) Although the “two server” data center may be possible for some levels of scale, things like grid computing will be very important. In that environment - does it make sense to have separate networks for storage, clustering and traditional networking or is there another opportunity here for convergence?

  5. Fabric convergence in the data center certainly seems to be a reality. Whether it is a derivative of the SCSI/Fiberchannel world such as Infiniband or 10 Gbps and up Ethernet with cut-through switching, bringing the packet, SAN, and clustering fabrics into a single multi-service converged fabric makes sense and will definitely happen. Fabric convergence eliminates interfaces, reduces footprint and eliminates power consumption. With the IEEE defining low-power Ethernet implementations in 802.3az, it is just a matter of time until we eliminate fabric duplication. Data center complexity and power are key issues of the next 5 years.

  6. As I have typed in the past, many times ownership of these networks is in vertical management silos. I also suspect there are some regulatory reasons such as Sarbanes-Oxley driving certain separations of concerns. I agree there are technical efficiencies to be had, but technical convergence by itself rarely succeeds without organizational convergence.

    So, look at the customer. If their management is vertically integrated and partitioned, guess what their network and data center looks like.

    I find the mainframe analogy both accurate and ironical.

  7. Management silos do come into play. But I am seeing evidence now that the storage team can allocate the storage, the computing team creates the VM and OS+app stack, the network team then provides an IP address and configures the load-balancer. It can and is happening on a converged infrastructure and is tied together with a workflow system. What is interesting here is automating this workflow - it really highlights both the silos and the opportunity to reduce operational expense. These silos can live over time unless there is a compelling business reason to break them down. SOX and other process controls such as separation of duty may actually keep the silos driving the workflow even though it is possible to take automation further.

    If you start to embrace the VM model, it is clear to me that de-coupling storage from the physical server and moving away from HBAs, local-storage and so on is the only way to go. This opens up the opportunity for fabric convergence even with the silos.

  8. You are preaching to the choir here.

    My point is that the ultimate compelling business reason is power and control. In many [most?] organizations, technical opportunities are subordinate to who controls the budget. Seeking out the organizations that are flat and understand the value and power of horizontal integration will make the optimizations you suggest succeed.

  9. Brad, what’s your opinion on cloud computing (IBM Blue Cloud, Amazon Elastic Computing. etc)? Do you think large organizations will be willing to move out of data centres altogether - assuming the services provide security and availability for their applications and information? It’s still very immature, but seems like the ultimate in computing resource sharing.

  10. This style of utility-computing looks very promising. The ability to rent capacity, especially large amounts of capacity for short periods of time should be another great way to reduce the costs associated with wasted/idle IT resources. I wonder to what degree large Enterprises will use this model. Will it be for specialized workloads? Could it make sense for “standard” line of business apps? Would it increase or actually decrease risk? For a number of reasons, I don’t think it will happen soon for large enterprises. A key question here is “where do you draw the line in terms of strategic advantage?”. At one extreme, it is possible to outsource everything: infrastructure, operations, application development and so on. Nicholas Carr (google “Does IT Matter?”) would argue that building your own Data Center is akin to building your own power plant. I’m not yet convinced that CIOs are ready to agree.

  11. A perfect example of the technical opportunities being subordinate to a vertical mangagement power structure. Without their own physical data center and large computing budget the CIO might not have the huge budget or as many wet noses reporting to them. That would make them not nearly as powerful and visible. Good luck selling that idea to the CIO :)

    For the enlightened company, I think cloud computing makes sense in the same way bandwidth sharing makes sense on a LAN for non buisness critical traffic like a “Red LAN, Blue LAN”, you would have “Red applications and Blue applications”

Leave a Reply