SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Mellanox Technologies, Ltd.
MLNX 124.890.0%Apr 27 5:00 PM EST

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
From: PaulAquino12/30/2016 4:25:47 PM
   of 954
 
Inspur server virtualization InCloud Sphere 4.5, open the "Super Mario" mode

Watchstor Anonymous Shu 2016-12-29 20:32

In the cloud on this high-speed road, with the widespread use of virtualization, hardware and software products iteration faster and faster, such as traditional Fast Ethernet, or even Gigabit Ethernet is difficult to meet the future requirements of the pool of resources, CPU introduced more New hardware-assisted chip and register functions and other new problems to the user problems, to prevent the formation of "butterfly effect", we need to stifle the problem in the bud.

OS InCloud Sphere is a core component of the IaaS layer that organically combines storage, networking, and other peripherals with server virtualization to make the entire IT environment more usable, scalable, and secure than individual physical hardware , For the enterprise core business cloud migration, build enterprise cloud data center provides basic technology and ability.

In order to better meet the needs of users and improve data center performance, Inspur released OS InCloud Sphere virtualization 4.5, based on the 4.0 version, the depth of hundreds of domestic users to tap the needs of a comprehensive upgrade product performance and ease of use .

Faster, more robust

Server virtualization system performance bottleneck mainly in the network and storage part. Traditional Hypervisor performance loss accounts for 5% -10%, performance loss is small, von Neumann architecture, to enhance the performance of more and more nodes rely on parallel processing technology, although the CPU process to 10nm below the technical and industrial difficulty is huge, Can enhance the space is limited, and GPU and other co-processor technology is developing rapidly. Therefore, ICS4.5 with the latest mainstream hardware environment, in the Hypervisor layer to do performance adaptation, and the maximum trigger parallel operation, to enhance the overall performance of network IO.

1. Network performance overall upgrade

The current 10 Gigabit Ethernet technology has become the mainstream, more and more users choose 40GB environment to deploy business networks, such as streaming media, large file transfer can not meet the requirements in the virtual environment, Ethernet card in the next 5 years will gradually enter the 40GB Era, is bound to virtual machine business network throughput has brought new challenges.

The Mellanox Connectx-3 40Gb / s environment test increases the throughput of InCloud Sphere from 29Gb / s to 100Gb / s by increasing the size of the authorized table lock and increasing the number of virtual CPUs in Dom0. S, performance up to 240%.



[Figure: three-port 40GB Ethernet card throughput test]

Due to process scheduling and network card cache and other structural design, resulting in physical network card performance is greatly reduced, how to run 10 Gigabit Ethernet environment and even 40GB Ethernet card capacity, Hypervisor needs to be addressed.

Bare metal architecture virtual machine bandwidth significantly improved, open 8-thread mode, the use of multi-queue card technology, InCloud Sphere throughput from 17Gb / s upgrade to 37Gb / s, fully parallel packet processing, close to the physical network card linear forwarding capabilities, performance Up to 120%.



[Figure: three-port 40GB Ethernet card bandwidth optimization test]

2.Windows type Docker support

Docker centralized management. Linux and Windows environment Docker unified management, extended Docker monitoring and management capabilities, integrated console and log management, real-time monitoring and diagnostic container system, reduce container deployment and management difficulty, unified management and control environment, lightweight isolation, can be lightweight isolation Configuration and Mobility. Single host supports more than 1000 + containers at the same time.



[Figure: Hypervisor on the Docker life cycle management]

3. GPU performance enhancement

Traditional GPU virtualization needs to configure the GPU graphics hardware, and the server PCI expansion slot length and height have a higher demand, often require dedicated graphics workstations. For the use of high-performance GPU scene, often using multi-tasking implementation of the IO and computing cluster environment for low-speed GPU virtualization scenarios, high cost graphics cards, it is difficult to batch use.

InCloud Sphere and Intel jointly developed to create a GPU without additional hardware with the vGPU solution: the first to support Intel GVT-g graphics virtualization technology, VM installed driver, each CPU supports 7 vGPU. , The use of standard Iris Pro Graphics driver compatibility, optimize the virtual machine display for low-load video encoding and decoding, rendering, game processing and other scenes.



In the environment for colleges and universities and research institutes, there is a wide range of scientific computing and in-depth learning needs. The number of vGPU sessions supported by a single InCloud Sphere host is increased from 96 blocks of 4.0 to 128 blocks.

• Scientific Computing: Scientific Computing and High-Performance Computing Co-Processors With NVIDIA's graphics card virtualization, ICS provides higher-density vGPU densities and CUDA toolkits are used to build GPU development environments for C and C ++ developers. The bottleneck of traditional high performance computing lies in computing nodes. The core contradiction lies in the granularity of task scheduling. If the task is too small, the number of clusters, message queue and data aggregation are high; if the task is too thick, it can not solve the problem of single GPU node physical performance . Therefore, the value of vGPU is to provide more coprocessors, sub-tasks for computing less demanding.

L depth of learning: multiple vGPU, SIMD (Single Instruction Multiple Data Stream) architecture, good for large quantities of data parallel processing. Such as cuDNN technology. Neural networks require a large number of matrix computing operations, such as autocoder, sparse coding trustworthiness network and so on.

More concise, easier to use, the private cloud also play flat

Full consideration of access flexibility and Linux and other user scenarios, 4.5 version of the web user interface to do a lot of optimization, the administrator does not accept the terminal environment, access is convenient, anytime, anywhere access and management. While compatible with the original client mode, and optimize the integration of the SSH console, only consume the bandwidth of the transmission instructions for the Linux class of client operating systems to provide more access and control methods.



[Figure: ICS 4.5 Web interface]

Automated Repair Capability Enhancements: Customers can use the Windows Update mechanism to install I / O drivers (PV drivers) that provide enhanced storage and network performance. Customers who wish to install the full version of the hypervisor management agent (performing VM lifecycle operations or monitoring performance) should also install the component in Tools ISO.

Enhanced Storage Migration Non-shared storage environments, virtual disk blocks are hot-migrated between different hosts, supporting power-on states, stop states, and suspend states. Therefore, it greatly facilitates the switching of services between different servers and storage environments. Virtual machines can be scheduled on demand in any environment to facilitate system expansion and hardware upgrade.

InCloud Sphere 4.5 is designed to give users a faster performance experience and user-friendly interaction. In the future, InCloud Sphere will continue to adhere to the product's stable and reliable, safe and easy to use continuous investment, to further enhance the performance, to provide users with stable and secure server virtualization system.
news.watchstor.com
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext