Neil Pound Pound argues that innovation is as much the province of users as it is of engineers and technologists. He then wonders why the UK public sector is not more adventurous in setting visionary goals to force the pace of innovation in the IT industry enabling more creative delivery of some public services.Read more...
Cloud computing promises increased efficiency for service providers and enterprise IT, and greater agility for users. To realize those promises, the cloud needs to be made up of large, shared pools of standardized and virtualized resources (including server, storage, and network elements). The larger the resource pool, the greater the efficiency, so scalability is key. But there is a problem. Most networks today currently don’t scale well. Read more...
I’ve talked about how today’s traditional network infrastructure resembles a tree and how much complexity and latency these legacy tree structures introduce.
These tree structures were designed primarily to handle the north-south traffic prevalent in the user environment; they were not designed to handle the east/west traffic of the modern data center. In order to go east and west, traffic must also travel north and south, which means it must pass through more switches. Each switch performs another Ethernet lookup, which adds latency. If switching can be contained at the access layer, it is pretty fast—from 1 to 5us. However, if east-west traffic also needs to run up the tree to the aggregation switch, a single transaction may now take 20-30us. If this traffic has to go to the top of the tree, the resulting latency is even more egregious—up to 100us to traverse a full data center network. In such an environment, it becomes more difficult to ensure consistently low application latency.
A more efficient architecture for modern applications would directly connect all processing and storage elements to each other in a flat, any-to-any network. “Flat” implies there is only a single address look up to process each packet, minimizing latency, while ”any-to-any” implies dedicated bandwidth to avoid congestion and maintain optimal latency. In such a world, all applications would enjoy optimal network performance. But is this too good to be true?
At Juniper, we believe that flat and any-to-any architectures are fundamental attributes of a “fabric,” and a fabric, not a tree structure, is the ideal topology for the data center. With Juniper’s Virtual Chassis technology, we apply the concept of a fabric across as many as 10 access switches, creating a flat environment for up to 480 servers. Virtual Chassis technology also eliminates the aggregation layer, flattening the overall layer 2 network from three tiers to two. Not only is the network now faster, it is also less expensive to own and operate through the elimination of an entire switching tier.
Looking forward, Juniper’s Stratus project will scale up the concept of a fabric and allow a single flat fabric to span the entire data center. This will further improve application performance while reducing the cost of the data center network. This is made possible because we have not built our business on propagating a legacy architecture and forcing the applications to conform; instead, we have built our business by using innovation to rethink the architecture and build a fabric that conforms to the needs of the application.
Keep the trees in your garden—don’t plant them in your data center.
Check out this animation to learn more:
Exploring the vision for the networking industry and the issues shaping its future.