Next gen data centre architecture Part 1: Juniper’s single switching fabric

| March 24, 2009 | 1 Comment

Juniper Networks’ Lam Chee KeongIn the past couple of weeks, cloud computing has really taken off, with both service providers, server makers, and switching vendors all making big announcements on service deployments and new gear designed for cloud operating environments.

Specifically, both Juniper and Cisco made significant announcements in their vision of a next generation data centre environment that can support the performance requirements of cloud computing services.

In this first article of a two part series on the two companies’ strategy and visions for cloud-based computing infrastructure , Green Telecom talks to Lam Chee Keong, regional enterprise solutions manager at Juniper Networks, about the thinking behind its Stratos Project and why there’s a need for a single fabric.

Green Telecom: So what is Juniper’s next generation data centre vision?

Lam Chee Keong: I think you have heard a lot about cloud computing and it is also common knowledge that cloud computing means different things to different people. One of the things that I think we can all more or less agreed on is basically that cloud computing is the ability to deliver services through the cloud. It is actually leveraging off on the public Internet infrastructure.

The whole idea of cloud computing is to be able to move to a point where computing can be treated like a utility, or what we called utility computing. With utility computing, you come to a point in time, like electricity. With Electricity, it is like utility for all of us. We go to anywhere, the office, home, or in any public place, there is a power point and I can plug my appliance or device into the power point and I turn it on and straight away I get power.

So cloud computing is moving in such a direction, to achieve a point where computing becomes a utility. Now cloud computing isn’t really new right, in a sense, you are delivering applications across the public infrastructure is not really new – it has been around for a while. With the new technologies, like Software as a Service, and novel concepts like this, I think is taking cloud computing to a whole new level. For enterprises, or end consumer or customer of cloud computing services, there certainly is a compelling reason with tangible benefits which are drivers for this.

So I think the two key reasons for this is first the issue of costs, especially in terms of applications – not only in acquiring the license and contracts, but I think with enterprises typically, when you acquire software applications, they typically overprovision or under-provision, this effectively hamper business agility, business objectives. If you overprovision, you basically waste costs. If you underprovision, you might not be quick enough to get additional licenses to meet business objectives.

The second driver is in the area of opex and management. With this utility as a computing concept, I don’t have to worry about, or invest in, headcount to manage the applications, to worry about upgrading, installing it, maintaining it, and basically all these, I can move, transfer, this kind of task, as well as risk, to the cloud provider, the service provider of the cloud. These are the key drivers for cloud computing.

And it is not just SaaS, but it basically comprises the applications, it can be storage services, security services, all kinds of services can be hosted in a cloud and made available to the public. So I think the other reason cloud computing has gained another level – one reason as I mentioned earlier was the way we are use applications and software, like SaaS, the second reason is the Internet infrastructure becoming more available with high performance networking capability. Today, broadband to the home, Internet access all the way to the home, is talking 8 Mbps, 10 Mbps of bandwidth, compared to years ago when you are talking about dial up. I think this is also fueling cloud computing considerations as well.

With cloud computing, on one end, you have users, on the other end, you have data centres, where applications are stored and located. So there is three main parts; the user and the devices, the infrastructure in between, which is usually the Internet, and on the other, there public or private data centres that hosts the applications or the services.

The user end, as I mentioned, you have high speed access to offices and homes, so it is kind of like a given now, high speed access. In the middle, the public infrastructure is also offers high performance networks, so that part has caught up as well.

Now the problem is the data centre at the other end. The problem with data centre today is that they are still based on legacy architecture, and there has not been much innovation as far as the network in the data centre is concerned. Of course, you may say, that is not true because there are new products and stuff for the data centers. Well sure, everybody is announcing new switches and routers in the data centre, but the problem is that data centre architecture is still based on a traditional, campus like – what we called a three tier concept of switches. So you might have very fast switches in the data centre, but the architecture needs to change to meet cloud computing environments.

For computing as a utility and cloud computing to really be transparent and ubiquitous to the user, I think the user experience is really important. If it’s going to take a long time to download an application, then the user experience is not good and the pick up rate will not be good.

So the data centre architecture is due for an overhaul. We have spoken to consultants, large public cloud providers, private cloud providers, and even our partners, systems integrators like IBM, who have built a lot of data centres, who understand the kind of requirements that goes into a data centre – that was how this Stratos Project on data centre fabric was actually born.

So how is it different from what is available today?

So the idea is that the data centre fabric that we are talking about with the Stratos Project and working on, brings a whole new meaning to the idea of the single, flat architecture for data centre. As I was talking about before, a lot of data centre today are still built based on the idea of a three tier, an edge layer for aggregation and a core switch layer kind of concept. In the past, without cloud computing, this was alright. But today, with the kind of data that is going across, especially if you are talking about converged data, you are talking about fibrechannel, infiniband, Ethernet, and lossless ability and all kinds of requirements, certainly having multiple tiers is a going to be a bottleneck and introducing lots of latency for this kind of application and data type.

So for Juniper, we think we have an opportunity to look at it and start from scratch, let’s take a fresh, blank piece of paper and let’s think about what is the most efficient and simplest way to implement the data centre network. So the idea is well: the ideal network infrastructure would be one single switch, where I interconnect everything. Today, a lot of infrastructures is not use to connect users to the servers, but interconnecting switches and other devices. So the simplest data centre infrastructure would be a single switch where you can plug in servers, users, storage and security devices, appliances. This would be simple, efficient and you don’t have to worry about multiple tiers.

The problem with a single switch, of course, is physically there are a lot of limitations, especially with the number of ports, you can’t scale. In cloud computing, we are staging, not just thousands, but tens of thousands of ports, you can have a switch with tens of thousands of ports. So scalability is a major issue with a single switch. You also have a single point of failure, so you have to make sure the switch is reliable.

So the idea is to have the concept of a single switch, so a logical single switch, but we actually expand it to a fabric. So all the intelligence is actually out in the edge, so you get the top of rack, or end-of-row box, and in between is actually interconnected like a fabric, in a way many-to-many type of interconnection. So with that, the whole fabric behaves like a single switch, and is managed like a single switch and it is a single tier. It is highly efficient and will be able to accept most converged data, like Ethernet, or fibrechannel, or even infiniband will work across it transparently. So that is essentially what our architecture would look like.

Is this applicable to existing installations? Obviously, if you are building a new data centre, then you and adopt some of this technology and implement it, but does this apply to existing facilities?

So there are a couple of things to think about. One of the reasons why we are announcing this Stratos Project so early because we expect this to be a multi-year project before we are able to release anything. The reason why we are announcing it is data centres have a long cycle. It takes a long time to plan, especially when we are talking about very fundamental things like the network architecture. So this allows our customers to plan ahead incorporate this when they need it. Secondly, we are not expecting this data centre fabric to appeal to everybody. So it won’t make sense in a campus environment. It won’t make sense in a small data centre environment.

However, when it comes to large data centres, we will appeal to them. The idea is that it will co-exist and it will work with the existing infrastructure without ripping everything out. In fact, it will augment and complement, for example, our existing EX line of switches perfectly.

So it is kind of like JUNOS network operating systems, so you can link up all the switches into a virtual fabric kind of thing?

That’s right.

So when do you expect some of the technologies from the Stratos Project to be released?

So the Stratos Project is something that is designed from the ground up using our expertise, experience, silicon design and testing, so we are expecting this to be a multi-year, at least a couple of years, effort. We are not expecting any revenue from the project this year.

In terms of the technologies that are being developed, like the 30 patents already submitted in the US, and how these will be commercialized, is there a roadmap?

Right now, I can’t comment on that right now unfortunately. I think as time goes on, we will continue to keep the public inform of what is happening.

In Part 2, Green Telecom looks at Cisco’s new Unified Computing concept.

Be Sociable, Share!

Tags: , , ,

Category: Cloud computing, Data centres, Featured articles, Green ICT, Interviews, Networks

Comments (1)

Trackback URL | Comments RSS Feed

Sites That Link to this Post

  1. A tale of two next gen data centre architectures : | March 25, 2009

Leave a Reply

%d bloggers like this: