Savvis on Cisco's Unified Computing System

Bryan Doerr, Chief Technology Officer, Savvis and I communicated about Cisco's Unifed Computing System (UCS). Here are his answers to some questions that were posed to him.

Bryan Doerr, Chief Technology Officer, Savvis and I communicated about Cisco's Unifed Computing System (UCS). Here are his answers to some questions that were posed to him.  Since Savvis currently operates 29 data centers (a total of 1.4 million square feet of space), his views could be seen as representing very large organizations as well as hosted and managed services providers to some extent. Read on to read more about Mr. Doerr's comments.

Here's what he had to say about Cisco's offering.

What products do you currently use?

Savvis is a relatively large volume consumer of enterprise servers, which we offer as a service to customers.  We traditionally evaluate and select servers from Dell, HP, and Sun for our various managed service offerings.  We try to limit our choices to minimize variability in our operations so we select one provider and revisit this decision occasionally.

Why are you considering a change?

Server technology and related management systems are evolving to support a new generation of deployment design which features virtualization as a key underpinning.  The traditional server, which was designed with a different deployment model in mind, may no longer be the most cost effective to operate.  The Cisco UCS may offer improvements in critical areas.

What tangible benefits do you expect to obtain through the use of UCS?

There are several reasons why a new server architecture might offer benefits.  First, by repackaging the server and network functions into a single system, new cost optimization is possible.  Secondly, by integrating Ethernet and Fiber Channel into a single data center fabric, cost and operational complexity can be improved.  Third, but natively supporting virtualization and introducing new quality of service controls, the goal of optimizing multi-tenant environments, as found in cloud offerings, can be achieved.  Finally, as a service provider, we operated both virtualized and non-virtualized environments and efficient support for both is important.

What advice would you offer others facing similar circumstances?

Consider how virtualization is impacting your entire infrastructure deployment environment and don’t be afraid to revisit fundamental tenants related to the building blocks of these environments.  Resource capacities, standards, business forces, and suppliers that guided the evolution of the current datacenter design patterns are changing and new options are emerging.  Cisco’s UCS is an example and may be appropriate for larger scale deployment environments with both virtualized and non-virtualized deployments.

What is your vision for how products like this will impact the market?

We are continuing to see decoupling from hardware in the evolution of applications.  Accordingly, the underlying hardware environment including servers, storage, and network are starting to evolve independently, driven by cost and power savings requirements.  Cisco’s UCS is an example of these new server and network systems.  Its impact will be determined by both the rate of conversion to alternative architectures by server buyers and by the rate at which other innovations in this area emerge.

What were the key attributes you were looking for in this beta program?

Our goal in participating in the beta was to assess the UCS for feature maturity, reliability, management system functions, power requirements, and performance.  Taken together these strongly affect the type of service we can offer and the cost to offer this service.  Our goal is evolve multi-tenant cloud offers to include a new level of security and resource guarantees and to find another generation of cost optimization in the systems used to offer these services.

If you did not use the product, what would your options be?

We would stick with conventional server technologies.

Snapshot analysis

Increasingly large and very large organizations are managing and configuring the industry standard systems found in their datacenters in a very similar way to that deployed by suppliers of managed or hosted services. So, the good Mr. Doerr's comments can be taken as a view of the future for many organizations. As with suppliers, such as Savvis, these organizations are seeking to create self-managed multi-tenant configurations that can support a large number of workloads that are isolated from one another through the use of virtualization technology.
Everything old is new again
Having lived through the age of mainframes and minicomputers, I'm not surprised to see the industry move back to some of the better features that these systems offered. Centralizing the equipment needed to bring a computing system to life had a number of benefits. It was typical at that time for the computers, I/O, networking and even storage to be in one large cabinet or group of cabinets. The technology available at that time forced hardware designers to consider distance's impact on latency more than today's systems.

As I recall, the VAX-11/780 processor, for example, was made up of over ten very large circuit cards. These cards, of course, couldn't be very far away from another if the machine was going to perform well.

Cisco, and all of the other suppliers of blade computing systems such as Dell, eGenera, HP, IBM, etc., have all started moving more and more functions back into the central system cabinet to minimize the impact of having each working solution end up being configured as a herd of individual boxes in a number of racks. The key difference is that each function is likely to be provided by a separate computer.

This means that each of the functions is being supported by a more intelligent, software controlled, device. This also means that the hardware supporting a function has been engineered to perform that function very well and at the lowest possible cost.

You'll note that Bryan always spoke about Cisco's UCS as something for the future. This is because the first few configurations have only recently been announced and are not in use in enough places for a datacenter's manager to have full confidence in them. Cisco's competitors, on the other hand, have years of experience in the field.

Thanks, Bryan, for taking the time to answer my questions!

Newsletters

You have been successfully signed up. To sign up for more newsletters or to manage your account, visit the Newsletter Subscription Center.
See All
See All