Cisco's Big Data Practice

In the SPF Conference, Cisco Greater China Data Center, chief architect Xie for large data solutions delivered a speech, SP hereby highlights the contents of the following.

Large data from the IT perspective will bring business drivers, will bring value-added services, this is the consensus. However, large data architectures differ greatly from traditional data center data management methods. These differences have created many new changes and challenges in the IT industry. What can Cisco do to help overcome these challenges?

In fact, the challenge is mainly focused on three areas. First, the traditional Hadoop approach has a very critical node, which is equivalent to the work engine, where all data management is operated through this engine. Failure of this node will make the entire large data system unusable, the business completely paralyzed. Therefore, it requires high availability.

Second: Hadoop usually has multiple copies of data to hold the data, and when a copy of the data fails the system will help replicate, which is implemented through the network. This can lead to a large amount of data replication in certain situations, especially when the distributed system is doing a lot more.

Third, data storage in the past is centralized storage connected to the server. The entire data management layer only needs to manage several nodes. For example, a traditional database has two nodes or three or four nodes, and in a Hadoop environment there will be dozens of or even thousands, and thousands of large clusters that are likely to span different data centers. This means that management points are very large and management costs are rising dramatically. These are the challenges that big data can face.

Cisco has a unified computing system for the above challenges. Future data center development, computing must be to the horizontal expansion. This means that virtualization, cloud computing and large data have a common feature of distributed development. A distributed system necessarily needs to be associated with the network, so Cisco wants to make a product that integrates computing and networking, which is Cisco's UCS server System.

The core of Cisco's server system depends on a pair of gigabit matrices, which are the management points of the computing domain, and all the servers hanging on the compute field are just a unit of computing, and it hangs on which applications are all determined by the network layer. Cisco's system itself realizes hardware virtualization, and the server's attributes are raised to the network, defined by the network.

Second, Cisco's system will make the management domain larger, as long as customers connect to the swap group, you can control all nodes, which will greatly shorten the management point.

Third, Cisco's system integrates two self-healing capabilities, how to achieve high availability, in two ways, Cisco offers hardware solutions. Since Cisco's system is AB two matrix loops, so once a loop has problems, B-ring can be solved immediately, second, Cisco's system to achieve the hardware drift ability, such as the server's CPU is broken, customers do not need to reload a server, Cisco's system can be its hardware properties "write" to another machine, can be used directly. This http://www.aliyun.com/zixun/aggregation/17183.html "> helps customers save a lot of software maintenance costs.

The development of large data now basically presents two trends, one is to insist on the independent open source way. Another is to provide their own solutions, in the original large data architecture to provide their own products. Both of these ways Cisco will provide good support.

To sum up, Cisco's unified Computing system, for key nodes, Cisco can provide hardware drift capability of the technology, attendance failure can be quickly repaired.

Second, it solves the problem of a wide range of management, assuming that 160 servers, it can do a very large data cluster, which means that there are 160 management points, using Cisco's system, managers only need to manage one of the points, can monitor all management points.

Finally, Cisco's system ensures its high availability and performance. High Availability is Cisco's two self-healing capabilities. It is self-evident that the benefits of fast and large data replication for nodes are obvious.

You might want to focus on another trend in big data, where many vendors have introduced new solutions based on their own technical features, and the hottest in the last year is SAP's Hadoop, which actually solves the problem of fast data processing through memory. SAP's Hana can be expanded in a number of ways by putting data in memory. It is embodied in Cisco Solutions, mainly divided into two parts, the first is a single node, generally small scale nodes. There is also an extensible node.

Cisco scalable node solutions that require local disk support, with two partners in storage, one for EMC and one for NetApp. Our solution integrates networks. Therefore, you do not need to consider the expansion of the connection module, but also do not need to rewiring.

Today, the main introduction of Cisco Computing products, Cisco Computing is not a simple calculation, is based on network development. In the future, especially in the more and more stressed distributed systems, Cisco Computing will bring a lot of surprises. (Weidong)

(editor: Heritage)