By guest authors Shaloo Shalini and Pablo Chacin
SM: What role do you see for a private cloud to have in your data center? Could it eventually become a substitute?
MS: When I think of private cloud, I think of more virtualization technology to get more utilization out of existing assets. That is journey or crusade most CIOs are on, even if they haven’t embraced the terminology “private cloud.” They are looking at the diversity of assets they have in their data centers due to acquisition of specific business applications. They are going back to their CFOs and telling them, “I don’t want to support five different types of hardware platforms, three different types of databases, five different kinds of operating systems. I want to get down to one or two platforms that I can manage and operate on a very reliable basis, and in the future when we embark on new acquisition of new application I want to make sure that I can work on one of these more standardized platforms.” With that kind of philosophy, people are standardizing to platforms like Wintel or Linux. They try to scale up by means of virtualization, so that they can achieve the economies of scale that you can achieve when you have lots of one-off platform in the data center.
SM: How do you distinguish between private and public cloud? The way we see public cloud is when some service provider hosts some application, and private cloud is when you host application on your own data center that you are provisioning for your own employee base. Does this definition gel with you or is there some nuance I am missing here?
MS: No, well, in general is right, but you can triangulate in on a private cloud along three different indices. The first question you can ask is – “Are the assets on my balance sheet and P&L? Do I own those assets?” Second would be location: “Are they sitting in my facility or in a facility I am leasing or own in some part?” Third would be from the management point of view” “Who is managing those assets?”
I don’t think you have to get a yes to all three of them to call it a private cloud. You can have your asset sitting on some one else’s data center floor that you are leasing or managed by vendor, and that would be one out of three, but to me that would still be a private cloud. There is a shade of gray. If it is managed under your supervision by a contractor or you put them on your data center floor and manage them, it is just being financed in a different kind of way. That is how I see it.
SM: Well, if you are drawing a line between capex and opex, if you are making capex investment then you would surely call the asset a private cloud, no?
MS: Yes, that would be the presumption. If you get three out three indices; that is, I own them, manage them, and they are in my data center, then they are definitely private cloud. But say you have some asset in a facility on lease you have leased administered by a vendor then to me. It is a private cloud? But yes, your definition is right; if you are paying for the asset then it is private cloud.
SM: So where are you in that configuration? I know you use some number of SaaS applications which are pure public clouds and some IaaS which are public clouds. What areas are you invested in that you would legitimately call private cloud?
MS: We have R&D laboratories in six different locations around the world: four in North America, one each in Israel and India. We view the server and storage assets in six different locations and manage it as a single cloud computing laboratory.
In the past two years we standardized on Wintel architecture. We have Windows operating system sitting on x86 architecture, and we have procured a significant number of servers from Dell that we have heavily virtualized.
In the course of doing that, we started with a total physical inventory of 15,000 machines in six locations. We can now deliver the same amount of horsepower to the number of servers through virtualization with a footprint of 5,500 virtualized servers. We achieved standard benefits; we were able to reduce our service sprawl and lower our power consumption. We were then able to cut some of our labor costs to support that kind of an infrastructure. Some of the real benefits of cloud, whether it is internal or external, are innovation and time to market.
So I tell people, one of the real benefits in our particular business is now folks that are doing our software engineering don’t have to negotiate with IT for over three to six weeks to get access to the assets for getting their real jobs done. If they adopt one of our standard configurations, we can provision that out of the pool in three to four hours on a given workday. It is hard to calculate the savings accurately. It is not as if the IT people were not doing anything for that three- to six-week period while the engineers were waiting for their assets. You know we basically take entropy, delay, and frustration away from the IT team so that they can do the real work that they want to do and provide more value to the business.
SM: Now, it sounds like you have done all the work to architect this private cloud. Do you see a scenario or a time when a more fully architect-ed solution will come from various vendors? I know they are working on it. I know IBM, Sun, HP, and even Cisco are trying to come up with virtualization architecture with optimization done and giving the end user a full solution as opposed to something that you have to architect yourself.
MS: They are basically selling you their stack. Sun, I think, at one time used the term “data center in a box” for what they were trying to do with servers.
Yes, I think people will try to do that. I wouldn’t predict that would be successful, nor would I necessarily vote against it. See, the undertow there is the historical legacy relationships you have with your existing vendors. There are lots of new companies showing up on the landscape. If they are going to be making their first major acquisitions of data center hardware, then I would think those pre-configured stacks would be very appealing. Inevitably, there are going to be legacy skill sets, legacy vendor relationships, and legacy applications which have been optimized to run on various operating systems, suiting certain kinds of hardware and so forth. So although some organizations might take a couple of our mission-critical applications and put them on standard architectures that we can scale at will, you know we can install more extensions to the cookbook set of hardware assets needed in particular reference architecture. They might do it on a piecemeal basis. There are probably too many legacy applications and too many legacy skill sets and frankly too much legacy hardware to write that all off overnight and standardize one of the stacks.
SM: So you think there is an exit barrier here?
MS: Well, the tables get turned back on the people building those stacks. I mean, from a pricing point of view they could make an irresistible proposition, but they are probably not interested in doing that. With most of the stacks being marketed today, well, frankly, it is actually more marketing than engineering that has gone into it. People are taking that the products that they are already selling and they kind of put them together and they call them a stack without necessarily optimizing it in terms of way they work together or the management tools which you use for the stack, and so forth. If they get through a couple of generations of their individual products which they trying to build up into the stack and they are nicely integrated and have great management capabilities, then it might be more appealing. So between pricing and technical integration and not just marketing integration, those could come to be a dominant force in the market. That is going to happen.
SM: Has this move of taking your organization toward a private cloud or your virtual data center approach changed the relationship between IT and the rest of the organization? Or is it just a technological change?
MS: It has significant consequences. We talked about IT delivering services to the business. We have really become a service provider. I think it used to be an impediment in the past, a source of frustration and bureaucracy that people had to contend with, to get their jobs done. What is followed now is more of an enabler and more of a partnership. It is actually helping those IT people to be more productive. We have different relationships with our procurement groups because under the private cloud concept, we believe in procuring capacity in advance of demand. As most of the systems are required project by project and justified application by application, you end up with a big mess, a crazy quilt with all those kinds of systems because of tight individual business applications. On the contrary, in the private cloud model, you begin to standardize from one or two reference architectures. It is really good operating in this kind of architecture and be able to save some money.
When you buy capacity in advance of request, in a virtualized environment, when the server requirements show up there is no negotiation. Instead of arguing over a Sun server or going to buy a piece of IBM equipment, you go with whatever you standardize, say Intel blade architecture, and you have couple of blades available.
SM: I see. Capacity has now become a commodity.
MS: Yes, to the point that now you buy them in advance. That is a very different mindset. There is no real business case. I know there are people in a company who will be more productive now than in incremental store or service model.
SM: How does the customer service aspect work here? As IT infrastructure provider, you are almost like a vendor to your organization, and now your entire organization of thousands of people is expecting some quality of customer service from you at a pretty large scale.
MS: What we have done here, is very common in the industry. We have identified some standard configurations within that – Wintel architecture. We provision that through automated processes. So, going to the negotiation step and talking of service mentality, we tried to identify some standardized end points that would satisfy 80% of demand. Well, 70% of demand that was very successful and efficient in terms of servicing our needs.
SM: Are you now running stuff like helpdesk software to tackle frequently asked questions?
MS: We use a service request management tool so that people can effectively construct their own requests. We use our tool called Atrium Orchestrator, which is a run-box scripting tool that is a cookbook of assets that they have just requested. Next, we use Bladelogic to configure the asset and put them under production status. It is a pre-automated end-to-end process. The helpdesk is something we don’t do if users stay under the standard configurations; we have a fairly automated process.