A Fusion Architecture for Cloud and Big Data

Contributor: Shay Khosrowshahi
Posted: 04/14/2016
questions
Rate this Article: 
Be the first!

The opening of Intel IDF2016 happened on April 13 in Shenzhen, where Intel displayed its latest progress in technology in smart hardware, internet of things, virtual reality and 5G, with the three conceptions for the future world at the core - awareness, internet of everything and computing extension.

According to Intel, through 5G and next generation network, high-speed wireless connection will be enabled for cloud computing, data services and other connected devices in the future, offering whole new intelligent services including self-driving car supporting smart route options, intelligent cities and connected medical services.

Fusion Architecture infrastructure will support new information services. During IDF, Inspur displayed its 2nd generation fusion architecture integrated solutions - InCloudRack, a flexible, extensible and manageable solution optimised for enterprise applications such as private cloud and hybrid cloud.

On the hardware level, this InCloudRack is structured with distributed network in place of the conventional tree network, therefore enabling the restructuring and convergence of network and other resources. The InCloud OS V4.0, a cloud computing platform software equipped, has further made pooling and automatic scheduling of computing, storage and network resources a reality.

Distributed Network Breaking Bottleneck in Scale-Out Network Communications

Reconfiguration of hardware, including network and storage, has effectively enhanced the efficiency of software definition, enabling better performance for InCloudRack in typical application scenarios such as big data and cloud computing.

In the three major technological fields, i.e. computing, storage and network, computing and storage, performance has made improvements than the progress in network technologies thanks to the development of processors and the application of SSD and flash memory. Besides, enormous network traffic generated by big data and cloud computing has made network a bottleneck of server performance improvement at the moment.

The InCloudRack, a conventional network card and switcher, are replaced by UNE (unified network engine), with each UNE equipped with 8 PCI-E interfaces of 25G internally and 8 Ethernet interfaces of 40G or 100G externally. A distributed network array is formed by UNEs, constituting a rack-level switch fabric to enable connection with the backbone network.

The network design of InCloudRack shows advantages over conventional networks, with minimum communication latency of 2.3 microseconds, as compared to conventional architecture where such latency exceeds 10 microseconds. The strength of 77% reduction of communication latency is especially visible in I/O-intensive service systems such as big data and cloud computing.

Hardware Reconfigured + Software Defined: SDN&NFV Function Demo Based on Distributed Network

On the hardware level, InCloudRack has enabled further decoupling, centralisation and pooling, providing better support for SDN. During IDF, Inspur demonstrated enabling SDN and NFV of InCloudRack with InCloud OS.

As well as enabling software-defined basic virtual network functions such as multi-tenant-based virtual 2-layer switch and virtual router, InCloud OS also supports software-defined virtual firewall, virtual load balancer and virtual NAT, while providing virtual network connection.

Universality "All in One"

Another key feature of InCloudRack is its universality net to perfection. Geared towards integrated delivery of large-scale private cloud deployment, this product supports OLTP (online transaction processing) and enterprise IT applications such as CRM and ERP, while providing an optimal system platform for virtualised integration, cloud computing and big data, among other applications.

InCloudRack is equipped with various types of computing units -2, 4 or 8, as well as other types of extension plans including storage, GPU and micro-server, fully integrating critical applications, non-critical applications and other applications in an "All in One" model, hence an overall rack server with the most diversified scaling plans in the business. In contrast, other fusion architecture plans usually require extra hardware devices to deploy application systems such as database and heterogeneous computing, etc.