HP服务器技术白皮书1
hp服务器技术白皮书1.doc
惠普ProLiant服务器技术白皮书随着关键业务应用对系统资源提出了越来越高的要求,提高系统带宽已经成为一个至关重要的问题。
在分析和评估了现有的系统结构之后,惠普公司确定了一种新的系统结构,它具有更好的性能、较高的处理器、高效的内存技术和I/O 扩展性及更高的带宽,能够满足日益复杂的事务处理、数据库、互联网/企业内部互联网和文件/打印应用的要求。
HP最新的内存保护技术由于互联网在诸如电信、金融等对存储要求非常高的行业上的应用越来越广泛,增大服务器容量已经成为必需满足的要求。
未来,互联网的使用将渗透到各个行业,因此提高数据中心(data center)的计算能力成了生产商和服务商都关心的问题。
目前,服务器厂商在他们生产的服务器上采用了更快的处理器和更大的内存,这带动了相关技术的不断发展变化。
当前有很多企业都通过裁减IT部门职员的办法来缩减开支。
在此情况下,服务器必须具备更高的容错能力,以减少设备检修所需的时间和花费。
要满足大规模应用的要求,服务器的制造商面临的问题就是在扩展内存的密度、增大内存的容量的同时,保证系统内存的可靠性。
惠普也面临同样的问题,解决的办法是利用3种容错内存保护技术(fault-tolerant memory protection technologies):在线备份内存、热插拔镜像内存和热插拔RAID内存。
以上提到的惠普AMP技术是系统可靠性的保证,用户可以根据自己对内存可靠性的要求自由选择系统配置,这将为其工作取得成功提供必要的基础。
介绍目前运行在工业标准服务器上的重要的商用软件所需的内存空间越来越大。
从趋势上看,新的操作系统可以支持更大的内存,同时服务器内存的容量也在不断扩大。
这些年来系统内存的可靠性越来越高,主要原因是有了更好的制作工艺和更新的内存保护技术例如ECC -它是由惠普首先在工业标准的服务器上采用的一项技术。
然而,随着存储元件密度的增加、服务器容量的加大,内存发生错误的几率也跟着增大了。
HPE ProliantDL360eGen8(668812-AA1)-服务器技术白皮书
HPProLiantDL360eGen8(668812-AA1)技术白皮书HP ProLiant DL360e Gen8(668812-AA1)重要参数声明:仅供参考,以当地实际销售信息为准•重要参数•售后服务•产品类别:机架式产品结构:1U•CPU型号:Xeon E5-2403标配CPU数量:1颗•内存类型:DDR3内存容量:4GB•硬盘接口类型:SAS/SATA/SSD标配硬盘容量:标配不提供•保修政策:全国联保,享受三包服务;质保期为3年(3年现场保修,7x24全天候响应,4小时到场)•客服电话:800-810-2058;400-610-2058(周一到周五:8:30-18:00(节假日休息))•具体内容:HP服务器保修于指定日期开始和结束:保修开始日期可以由产品制造期间存储于产品中的日期、安装日期或者客户购买凭证上的日期来确定。
不同产品的保修时间可能会有所差异。
影响保修条款的因素包括销售产品的国家/地区,以及产品的类型。
具体的保修期以用户手册及HP根据该产品型号、序列号的查询结果为准。
进入官网>>详细参数•基本参数•处理器•主板•内存•存储•网络•管理及其它•电源性能•保修信息基本参数产品类别机架式纠错产品结构1U纠错处理器CPU类型Intel 至强E5-2400纠错CPU型号Xeon E5-2403纠错CPU频率 1.8GHz纠错标配CPU数量1颗纠错最大CPU数量1颗纠错制程工艺32nm纠错三级缓存10MB纠错总线规格QPI 6.4GT/s纠错CPU核心四核纠错CPU线程数四线程纠错主板扩展槽1×PCIe Gen3 (x16FH-HL),1×PCIe Gen2 (x8LP)纠错内存内存类型DDR3纠错内存容量4GB纠错内存插槽数量12纠错最大内存容量384GB纠错存储硬盘接口类型SAS/SATA/SSD纠错标配硬盘容量标配不提供纠错RAID模式1个智能阵列B120i SATA RAID纠错光驱标配不提供纠错网络网络控制器1个1Gb 366i 以太网适配器(4端口)纠错管理及其它系统管理iLO Management Engine、Insight Control 纠错电源性能电源类型热插拔冗余电源纠错电源数量1个纠错电源功率460W纠错保修信息保修政策全国联保,享受三包服务纠错质保时间3年纠错质保备注3年现场保修,7x24全天候响应,4小时到场纠错客服电话800-810-2058;400-610-2058纠错电话备注周一到周五:8:30-18:00(节假日休息)纠错详细内容HP服务器保修于指定日期开始和结束:保修开始日期可以由产品制造期间存储于产品中的日期、安装日期或者客户购买凭证上的日期来确定。
HP Integrity Blade Server系列产品技术白皮书说明书
Blades (BL860c i2, BL870c i2, andBL890c i2)World’s first scale-up blades built on the industry’s #1 bladeinfrastructureTechnical White PaperTable of contentsExecutive Summary (2)Product Family—A starting point (2)HP BladeSystem c-Class Enclosure Solutions and Integrity Server Blades (4)Building a Bladed Scalable System—Foundation Blade and the BL860c i2 (5)HP Integrity BL860c i2 Server Blade (5)Scalable Blade Architecture—Building larger systems (7)Inter Blade QPI Fabric Communication (7)Integrity Blade Link (7)HP Manageability Architecture (8)QPI Fabric and System Topology (8)HP Integrity BL870c i2 Server Blade (8)HP Integrity BL890c i2 Server Blade (9)Memory Architecture (10)HP Integrity IO Subsystem (10)Flex-10 and Virtual Connect (11)Conclusion (12)Executive SummaryThe HP Integrity server blade architecture builds on the best of HP blade technology with new levels ofscalability, spanning from the BL860c i2 single blade entry solution up to the BL890c i2 quad blademid-range offering. Benefits of the HP integrity server blade solutions include scalability, ease of use,flexibility, best-in-class I/O solutions, and total cost of ownership (TCO). This provides a compellingmission-critical Converged Infrastructure strategy for Integrity server blades.If you are considering upgrading from a current bladed or non-bladed server design to a nextgeneration server, this white paper is intended to highlight some of the capabilities of the HP Integrityserver blades to aid in your evaluation.Product Family—A starting pointThe HP Integrity server blades, as shown in Figure 1, are designed to provide a range of capabilities,replacing the prior Integrity generation BL860c, BL870c, and racked 4-processor module and8-processor module rack mount servers. The Integrity server blades product line is composed of theBL860c i2 single blade server, BL870c i2 two-blade server, and BL890c i2 four-blade server. Theprocessor, Memory, and I/O resources of the Integrity server blades can be adjusted, that is scaled,to meet application requirements. Integrity server blades enable compute, memory, and I/O capacityto scale across the product line, as shown in Table 1. In essence, the supported resources of theBL870c i2 are double those of the BL860c i2, and the BL890c i2 resources are double those of theBL870c i2.A complementary part of the Integrity server blades story is that the Integrity blade productsseamlessly coexist with other HP BladeSystem solutions within the c-Class enclosure, enabling mixedHP ProLiant and Integrity, and StorageWorks storage blade solutions within the same enclosure. TheIntegrity server blades and enclosure solutions are supported by the HP BladeSystem managementsuite of products.The HP Integrity server blade family will support future enhanced capabilities, including memory,processor enhancements, operating system offerings, partitioning abilities, and field product upgradesto further extend its capabilities.2Figure 1: Integrity Server Blades Product Family3Table 1: Blade CapabilitiesIntegrity BL860c i2 BL870c i2 BL890c i2 Processor module Intel® Itanium 9300 processor seriesProcessor/Cores/Threads 2P/8C/16T 4P/16C/32T 8P/32C/64T Chipset Intel® E7500 Boxboro/Millbrook ChipsetMemoryIndustry Standard DDR3 24 DIMM slots96 GB (w/4 GB)192 GB (w/8 GB)*48 DIMM slots192 GB (w/4 GB)384 GB (w/8 GB)*96 DIMM slots384 GB (w/4 GB)768 GB (w/8 GB)*Internal Storage 2 hot swap SAS HDDsHW RAID 0, 1* orMirrorDisk/UX 4 hot swap SAS HDDsHW RAID 0, 1* orMirrorDisk/UX8 hot swap SAS HDDsHW RAID 0, 1* orMirrorDisk/UXNetworking (built-in IO solution) 4 @ 10 GbE NICsFlex-10 capability8 @ 10 GbE NICsFlex-10 capability16 @ 10 GbE NICsFlex-10 capabilityMezzanine IO Slots 3 PCIe slots 6 PCIe slots 12 PCIe slotsManagement Integrity iLO 3 (Integrity Integrated Lights-Out 3), iLO 3 Advanced Packlicense is includedSupported OperatingSystemsHPUX 11i v3, OpenVMS v8.4*, Windows Server 2008 R2** (Future Support)Future enhancements include 16 GB DIMM Memory, Product field Upgrades and Partition support.Integrity server blades have an important set of reliability, availably and serviceability (RAS) features,provided in all key areas of the architecture. Capabilities such as, Intel Cache Safe Technology®,error hardened latches, register store engine, memory protection keys, double device data correction,and CPU sparing and migration as well as Advanced Machine Check Architecture (AMCA) forco-ordinated error handling across the hardware, firmware, and operating systems. RAS details arecontained in the “Technologies in the HP Integrity Server Blades” white paper.Virtual Connect Flex-10 features significantly expand the I/O capabilities of the Integrity server bladesand will be described in more detail later in this white paper.HP BladeSystem c-Class Enclosure Solutions and Integrity Server BladesIntegrity server blades may reside in the c7000 or c3000 BladeSystem enclosures.The HP BladeSystem c-Class enclosure story is composed of four key themes:1.Simple to manage and easy to control: Onboard Administrator (OA), Integrity HP IntegrityIntegrated Lights-Out (iLO 3) Manageability, and HP Insight Control Environment2.Agility: Pre-wired and pre-configured make adding a new server simple3.Reduced capital and operating costs: Uses less power and less than half of the space comparedwith racked solutions.4.Ease of Integration: Flexible power configuration, same management tools as HP rack-mountservers4These four elements are applicable to Integrity server blades and become even more important as the size and capabilities of the servers increase.Table 2 details the capacity of the c3000 and c7000 enclosures for Integrity server blades.Table 2: Enclosure Capacity for the Integrity Server Blade FamilyEnclosure CapacityIntegrity server bladesBL860c i2 BL870c i2 BL890c i24 servers 2 servers 1 serverc3000(full height—4 bays)8 servers 4 servers 2 serversc7000(full height—8 bays)Building a Bladed Scalable System—Foundation Blade and the BL860c i2All Integrity server blades are built on the same basic hardware structure. With the Intel QuickPath Interconnect (QPI) fabric, Integrity Blade Link and the extensible HP Integrity iLO 3 solution, HP defines a family of servers, scaling from the BL860c i2 two-processor module system all the way up to theBL890c i2 eight-processor module system.The Integrity server blades are comprised of one to four physical blades, depending on the Integrity server blade product. Multi-blade solutions are regarded as being “conjoined” or interdependently linked together. The Integrity server blades design relies on a common foundation blade design that is replicated and configured based on the location, for example node address, within the QPI fabric topology, resulting in a scalable architecture. The Integrity server blade architecture can easily be increased in resources, for example processor modules, memory and I/O, with sufficient allocated QPI fabric bandwidth and I/O bandwidth to enable a balanced system as I/O resources are added. As the Integrity server blade architecture is increased or scaled from a minimum configuration to a maximum configuration, all management resources, tools, and user interfaces behave consistently, simplifying administration of the entire product line.HP Integrity BL860c i2 Server BladeThe foundation blade structure provides a set of blade features as shown in Figure 2. A foundation blade combined with a BL860c i2 Integrity Blade Link, will be configured to a BL860c i2 blade.The BL860c i2 server blade supports two Intel® Itanium® processor 9300 series (quad-core ordual-core) processor modules. The two processors are linked together by a full width QPI link providing 19.2 GB/s of bandwidth. Each Itanium processor 9300 series module contains two memory controllers with an aggregate bandwidth of 28.8 GB/s. Each Processor Module supports up to 12 DDR3 memory DIMMs, or 24 memory DIMMs in total for the blade. The blade will support a memory capacity of 384 GB when 16 GB DIMMs are available.The I/O subsystem is composed of built-in I/O functions and three I/O expansion slots. The I/O expansion slots are provisioned with x8 PCIe Gen2 links back into the IO Hub controller (IOH). This provides a raw total aggregate bandwidth of 10 GB/s per I/O Mezzanine card (send and receive).5The BL860c i2 server blade also has four 10 GbE Flex-10 NIC ports provided by two embedded dualLAN controllers. Each dual LAN controller is provisioned with x4 PCIe Gen2 links, providing a rawaggregate bandwidth of 5 GB/s (send and receive).The BL860c i2 server blade contains an on-board SAS RAID controller and a separate PCIe Gen1 x4path to an adjacent blade enclosure bay for support of partner blade SAS mass storage and tapebackup options. A manageability subsystem also resides on the blade which works independently ona single blade basis or as an orchestrated manageability subsystem if multiple blades are conjoinedtogether. As with all c-Class Integrity and HP ProLiant blades, the HP Integrity server blades provide aSUV (serial, USB, and video) port on the front of the blade for service and support.As shown in Figure 2, each Itanium processor 9300 series module provides three QPI fabric links tosupport conjoining of blades, using the Integrity Blade Link. This architectural approach enables thefuture ability to upgrade the BL860c i2 product to a BL870c i2 product, resulting in the scaling up ofinterconnected resources by 2x. The BL870c i2 product could also be upgraded to a BL890c i2,resulting in a further doubling of resources.Figure 2:Foundation Blade Architecture DiagramThe BL860c i2 server blade is designed to be socket compatible with the nextgeneration of Itanium processor modules as well.6Table 3:Integrity Server Blades Supported ProcessorsFuture ItaniumFeature Intel® Itanium® processor 9300seriesCores per processor module 4 EnhancedMulti-Threading Support Yes YesUp to 24 MB L3 Enhanced Processor module Highest LevelCache SizeDIMMs per processor module 2-12 2-12DRAM Technology DDR3 DDR3 Scalable Blade Architecture—Building larger systemsThe Integrity server blade architecture is based on distributed shared memory (DSM) architecture, also known as cache coherent Non-Unified Memory Architecture (ccNUMA). The Integrity server blade architecture utilizes the Intel Quick Path Interconnect (QPI) point-to-point fabric to connect the primary system components. Each node in the DSM architecture has access to shared memory. To maintain coherency between all caches in the system, a cache coherency protocol is implemented over QPI. The cache coherency spans all the distributed caches and memory controllers.Inter Blade QPI Fabric CommunicationInter Blade communication is achieved through a glueless structure composed of the Integrity Blade Link, HP Manageability architecture and QPI fabric topology.Integrity Blade LinkFor multi-blade products, for example BL870c i2 and BL890c i2, the Integrity Blade Link conjoins or ties together all the associated blades by providing a physical inter-blade path for the QPI fabric. Additionally, the Blade Link provides a path for auxiliary signals. An additional feature of the Blade Link is to provide description information for the manageability architecture, for use in configuration and initialization of the Integrity server blade hardware.Initially, when the blades are inserted into the enclosure, only the manageability subsystem will be powered on. The attachment of the Integrity Blade Link is required to conjoin the blades into a server, enabling full power-on and boot.The Blade Link physically attaches to the front of one or more blades and is unique for the BL860c i2, BL870c i2, and BL890c i2. The suffix of the Blade Link indicates the size of Blade Link, for example IBL-2 refers to a Blade Link which will support a conjoined two blade solution.7Figure 3: Integrity Blade Link Example—BL890c i2HP Manageability ArchitectureThe HP Integrity HP Integrity Integrated Lights-Out (iLO 3) Manageability system is an important aspectof the overall Integrity blades scalable architecture. From a user point of view, the Integrity iLO 3access point for single and conjoined blades provides a manageability console as well as commandand configuration abilities. As the product scales to larger sizes, the Integrity iLO 3’s on each bladescale as well, working together to manage the product, but providing the administrator with a singleIntegrity iLO 3 server user interface for simplicity. The enclosure Onboard Administrator (OA) behavesin a consistent manner with Integrity server blades and other blade hardware and tools. From anarchitectural point of view, the HP Manageability architecture is the back bone of the system, givingthe ability for each blade to be configured, initialized, and then connected into a scalable set ofnodes with distributed shared memory. Every blade ships with an iLO 3 Advanced license factoryinstalled to enable advanced remote management features such as virtual media.QPI Fabric and System TopologyThe HP Integrity server blades are designed to be scalable. The BL890c i2 supports up to8 processor module sockets, all accessible to each other through the low-latency high bandwidthsystem QPI fabric. Each of these processor sockets supports the Intel® Itanium® processor 9300series module.HP Integrity BL870c i2 Server BladeBuilding a scalable system using the Foundation blade architecture shown in Figure 2, two foundationblades conjoined together with a BL870c i2 Blade Link yields the BL870c i2 topology shown inFigure 4. The BL870c i2 is a four process module system with a fully interconnected QPI fabricbetween all four processor modules.8Figure 4: QPI Fabric, BL870c i2HP Integrity BL890c i2 Server BladeThe HP Integrity architecture can be scaled up to a larger system beyond the BL870c i2 server blade, using the foundation blade architecture shown in Figure 2. Four foundation blades may be gathered or conjoined together using a Blade Link to create the Inter-blade topology structure as shown in Figure 5, resulting in the HP Integrity BL890c i2 architecture.Figure 5:QPI Fabric, BL890c i29Memory ArchitectureccNUMA stands for “cache coherent Non Unified Memory Access.”For ccNUMA systems, CPU memory access time is related to the processor to memory locationproximity. For example a given processor module can access its locally attached faster than memorythat is attached to another processor. Stated in other words, processor module local memory, alsoknown as SLM (socket local memory) has lower latency than accessing memory connected to anotherprocessor module’s memory controllers by sending memory requests through the QPI link across oneor two nodes. Referring to Figure 5, a memory reference from one scalable node to an adjacentscalable node requires communication between the two nodes to access the memory at the target.With the Integrity server blades architecture, QPI fabric provides low latency and high bandwidth tofacilitate node to node communication.UMA or “Unified Memory Access” is a memory configuration in which all processor modules haveuniform memory access latencies. The memory interleaving terminology for this configuration is ILM(interleaved memory). While it is possible for any processor module to communicate with any otherwithin the server, an ILM configuration interleaves across either the top or bottom half of the cube ofFigure 5, resulting in only one node hop for a memory access.A feature of the Integrity server blades architecture is that the user may specify an optimal memoryinterleaving scheme that will be best suited to the user’s application and the host operating system.Different Operating Systems may prefer different configurations for better performance. The user canspecify a range of options from full SLM to full ILM. The default interleaving setting at EFI is “MostlyNUMA” with 7/8th of the memory configured as SLM and 1/8th as ILM. For further details on the topicof memory configuration and optimization please refer to the “Memory Subsystem Information for HPIntegrity Server Blades” white paper.HP Integrity IO SubsystemThe Integrity server blade I/O subsystem brings a broad range of capabilities to the HP Integrityserver product line. As noted earlier in this white paper, the I/O subsystem is supported by a lowlatency high bandwidth QPI fabric. Within the I/O subsystem, four categories of I/O are provided ona foundation blade:1.Three general I/O mezzanine card slots–PCIe Gen2 x8 for each slot–VC Flex-10 capable2.Four 10 GbE built-in LAN ports–Two dual-port NICs, PCIe Gen2 x4–VC Flex-10 capable3.One built-in SAS RAID capable controller–Supporting two hot plug SAS drives–PCIe Gen2 x14.ICH Mezzanine built-in I/O capabilities–PCIe link for adjacent enclosure bay partner blade support–USB controller–VGA controller–PCIe Gen1 x410As the Foundation blade architecture is conjoined to scale up to larger size servers, the I/O subsystem capacity scales linearly. For example the BL870c i2 provides double the capacity of the BL860c i2.For further details please refer to the “Technologies in the HP Integrity Server Blades” white paper. Flex-10 and Virtual ConnectThe Virtual Connect Flex-10 capabilities of HP Integrity server blades set it apart from its competitors. It provides up to four virtual NICs over a 10 GbE LAN server connection, and bandwidth limits can be dynamically configured on each NIC. Virtual Connect Flex-10 can also reduce the number of Ethernet mezzanine cards and blade interconnect modules required within the Blade Enclosure. Note that it is important to scale up the memory subsystem size as the I/O subsystem is scaled up. Please refer to the “Memory Subsystem Information for HP Integrity Server Blades” white paper for further details.The Integrity BL890c i2 server blade configuration example, using Virtual Connect Flex-10, allows the system to expand from a physical implementation of 32 physical NIC ports into a maximum of 128 virtual NIC ports.Table 4:Flex-10 LAN CapacityBase Flex-10 LAN port Configuration Maximum Flex-10 LAN port ConfigurationPhysical 10 GbE built-in NIC port count Virtual ConnectFlex-10 portcountPhysical 10 GbEbuilt-in NIC portcountOptional I/O10 GbE NICcard count(physicalport count)VirtualConnectFlex-10 portcountBL860ci24 16 4 2,(4)32 BL870ci28 32 8 4,(8)64 BL890ci216 64 16 8,(16)128The maximum Flex-10 configuration, as shown in Table 4, leaves one general I/O mezzanine slot per blade within the server available to support an addition type of I/O interconnect, such as dual port8 Gb Fibre Channel. A BL890c i2 System configuration example, customized for Storage Area Network (SAN), would support up to 12 2-port 8 Gb Fibre Channel cards, while still having 16 physical 10 GbE NICs that may be expanded to 64 virtual ports.Other I/O configurations can be implemented to provide the best combination of Fibre Channel, Ethernet and SAS or other types of I/O to meet the user requirements.11ConclusionHP Integrity server blades provide:•A broad product offering within the current c-Class c7000 and c3000 enclosures•Memory, storage, and I/O that scales linearly with compute power•Improved total cost of ownership by increasing Integrity blade features within the existingc-Class enclosure.•Consistent management and tools with existing c-Class blades•Support for Integrity, ProLiant, and StorageWorks storage blades in the same enclosure.With the rich set of features and choices, enabled for use within both the c7000 and c3000 bladeenclosures, HP Integrity server blades provide a balanced, scalable system, for single blade andmulti-blade conjoined configurations. The HP Integrity server blade architecture utilizes a low latencyhigh bandwidth QPI fabric to tie together each of the nodes within the server. The architecture alsoprovides a best in class I/O architecture, utilizing Flex-10 and Virtual Connect.Share with colleagues© Copyright 2010 Hewlett-Packard Development Company, L.P. The information contained herein is subject tochange without notice. The only warranties for HP products and services are set forth in the express warrantystatements accompanying such products and services. Nothing herein should be construed as constituting anadditional warranty. HP shall not be liable for technical or editorial errors or omissions contained herein.Intel and Itanium logo are trademarks of Intel Corporation in the U.S. and other countries.4AA1-1295ENW, Created April 2010。
华为VME+H.264 HP双核处理技术白皮书1.1
HUAWEI VideoConference WhitePaperVME+H.264HP : 为您提供“双核”视频体验1.概述高清视频通信的广泛应用及其背后的巨大市场前景,积极推动视频压缩、传输技术的发展。
从标清到高清,从高清到低带宽高清,华为依托强大的研发能力及技术优势,长期专注的视讯发展定位,持续为客户提供最优体验,最具性价比的视讯解决方案。
目前视频通信业界普遍采用H.264 Baseline Profile编码,主要基于其算法简单,硬件实现成本低,但在现有的网络环境中,使用H.264 BP构造高清(HD)视频会议系统所需要的带宽及网络适应性也一直为客户所关注。
H.264 HP(High Profile)编码标准可以有效解决上述问题,华为从未停止对H.264 HP的产品化研究,一直致力于为客户提供最具性价比的H.264 HP解决方案,并最终推动整个视频通信领域发展。
在未来的视频会议场景中,每个屏幕上的图像分辨率至少为1920×1080,假设帧频为30帧/秒,采用4:2:0格式,未经压缩前的码率可以达到746Mbit/s,如果直接在现有的信道中传输,需要有相当大的带宽,为了能在较低带宽下达到使用效果,且不丢帧,则必须对图像进行高度压缩,必然会带来图像的高损伤。
后处理增强技术就是为了消除在高损伤视频压缩下图像的负面效果,采用自适应滤波及清晰化的方法,来改善画面质量、提升观赏质量和视觉效果。
本白皮书概括性的阐述华为视频H.264 HP解决方案,及配套华为自主知识产权的创新性视频处理技术VME(Video Motion Enhance)。
1.1.H.264 High Profile编码H.264是ITU-T的VCEG(DS视频编码专家组)和ISO/IEC的MPEG(活动图像编码专家组)的联合视频组(JVT:joint video team)开发的一个新的数字视频编码标准。
H.264有四种画质级别,分别是BP、EP、MP、HP:1)BP-Baseline Profile:基本画质。
- 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。
惠普ProLiant服务器技术白皮书随着关键业务应用对系统资源提出了越来越高的要求,提高系统带宽已经成为一个至关重要的问题。
在分析和评估了现有的系统结构之后,惠普公司确定了一种新的系统结构,它具有更好的性能、较高的处理器、高效的内存技术和 I/O 扩展性及更高的带宽,能够满足日益复杂的事务处理、数据库、互联网/企业内部互联网和文件/打印应用的要求。
HP最新的内存保护技术由于互联网在诸如电信、金融等对存储要求非常高的行业上的应用越来越广泛,增大服务器容量已经成为必需满足的要求。
未来,互联网的使用将渗透到各个行业,因此提高数据中心(data center)的计算能力成了生产商和服务商都关心的问题。
目前,服务器厂商在他们生产的服务器上采用了更快的处理器和更大的内存,这带动了相关技术的不断发展变化。
当前有很多企业都通过裁减IT部门职员的办法来缩减开支。
在此情况下,服务器必须具备更高的容错能力,以减少设备检修所需的时间和花费。
要满足大规模应用的要求,服务器的制造商面临的问题就是在扩展内存的密度、增大内存的容量的同时,保证系统内存的可靠性。
惠普也面临同样的问题,解决的办法是利用3种容错内存保护技术(fault-tolerant memory protection technologies):在线备份内存、热插拔镜像内存和热插拔RAID内存。
以上提到的惠普AMP技术是系统可靠性的保证,用户可以根据自己对内存可靠性的要求自由选择系统配置,这将为其工作取得成功提供必要的基础。
介绍目前运行在工业标准服务器上的重要的商用软件所需的内存空间越来越大。
从趋势上看,新的操作系统可以支持更大的内存,同时服务器内存的容量也在不断扩大。
这些年来系统内存的可靠性越来越高,主要原因是有了更好的制作工艺和更新的内存保护技术例如ECC -它是由惠普首先在工业标准的服务器上采用的一项技术。
然而,随着存储元件密度的增加、服务器容量的加大,内存发生错误的几率也跟着增大了。
而内存一旦出现错误会导致数据无法使用,甚至系统死机,这会给商业活动带来很大损失。
为了确保内存的可靠性,惠普开发了3级AMP技术,它能够增加内存的容错能力,能够满足那些对实效性要求很高的应用软件的要求:用户可以根据自己对内存可靠性的要求自由选择系统配置-在线备份内存、热插拔镜像内存和惠普热插拔RAID内存(工业标准DIMM的冗余阵列)。
在这篇文章中,首先解释服务器发生内存错误的原因并说明为什么出现内存错误的可能性会增加。
接着详细介绍惠普采用的检查内存错误的方法的原理和局限性。
最后,介绍惠普ProLiant 300, 500和 700 系列服务器采用的容错AMP技术。
内存错误服务器中使用的内存模块是电子存储设备;因此很容易出现存储错误。
计算机中使用的存储设备有两种类型- 静态随机存储器(SRAM) 和动态RAM (DRAM)。
其中SRAM 做为缓存使用,这是因为它的速度快并且可以在关闭电源前一直保存其中的数据。
DRAM 芯片装在168脚的DIMM( dual inline memory modules )上。
每一个DRAM 芯片以电容行或电容列存储数据(即存储单元),这些存储单元必须不停地进行充电或者更新,否则其中的数据会丢失。
一个充电的电容器表示数据"1",一个放电的电容器表示数据"0" 。
充电和放电是由存储设备的电压决定的。
在进行读操作的时候,电容器的电荷水平决定数据是被读出为“1”还是读出为“0”。
例如,在电压为5伏的系统中,传感器将电压为+5伏特的电容器读出为“1”,将电压为0伏特的电容器读出为“0”。
只要电压更接近+5 V 而不是0 V,传感器就可以正确读数。
然而,如果电容器中的电荷受到外界的影响发生变化的话,读数就会不准确。
当服务器上运行的是重要的商用程序时,此类内存错误将导致运行中断,商业数据的丢失。
内存错误根据被影响的比特位数不同分成单比特错误和多比特错误。
HP防止出现内存错误的方法有两条途径可以防止出现内存错误:测试(testing)和使用错误检查/纠正技术。
一直以来,惠普是工业标准服务器内存品质测试方面的专家(惠普为其产品提供质量保证就可以证明这一点),同时它还在工业标准服务器内存的错误检查/纠正技术方面处于领先地位,并在AMP技术领域继续保持了这种地位。
先进的测试方法可以提高内存的可靠性内存芯片速度更快、设计更复杂,这使得对芯片进行测试变得越来越困难,花费也跟着变大了。
内存设备的生产商在测试系统上投入了很多的资金,并不断地改进其测试的方法、步骤从而确保设备的质量。
惠普通过不断改进内存的设计和制作工艺,将发生硬错误的可能性降到最低。
除了在制作上严格把关外,惠普更是对服务器上使用的每一个内存都做了彻底的测试。
而且测试范围不仅包括目前正在惠普服务器上使用的各个生产商生产的各种型号的内存,还包括每一款惠普新研制出的服务器内存,在惠普服务器采用新的处理器时,惠普也会对原有的内存进行重新评估。
只有通过严格的测试才能保证内存可靠性的不断提高。
有了对其产品质量的信心,惠普为它生产的内存提供3年的质量保证。
有了这个保证,只要是出现问题的惠普DIMM(其错误值超过了预先设定的错误极限),用户都可以向惠普要求更换。
服务器会记录DIMM出现的错误,而且惠普的管理软件(Hp Insight Manager)和诊断程序都可以对发生的错误进行确认。
错误检查/纠正技术惠普通过严格管理内存的加工制作过程和不断地对其内存产品进行测试,最大程度地降低了由于产品质量而导致内存发生错误的可能性。
但由于内存容量的扩大,发生内存错误的几率也在增加。
错误检查/纠正技术很重要,没有它,正在运行的重要商业程序会在没有任何告警的情况下出现中断。
这虽然是偶然现象但后果却很严重。
尽管采取各种方法避免硬件出现问题,但由于一些和内存不直接相关的错误的存在,内存错误仍然不可避免。
唯一可以真正改变这一状况的方法是:使用内存检查/纠正协议。
有一些协议只可以检查到错误,另一些却可以在检查到错误的同时把错误改正过来。
ECC 内存惠普1993把带纠错编码技术(ECC)的内存首次应用到工业标准的服务器中,大大降低了内存发生“致命”错误的几率,惠普是第一个这样做的公司。
ECC内存已经成为所有惠普ProLiant系列服务器的标配。
ECC技术优于奇偶校验(parity checking)。
奇偶校验只能做到单比特错误的检查,不能对检查到的错误进行纠正,且无法处理多比特错误。
但ECC 技术不仅可以检查到单比特错误,也可以检查到发生的多比特错误并能对单比特错误进行纠正。
它的原理如下:ECC将信息进行8比特位的编码,采用这种方式可以恢复1比特的错误。
每一次数据写入内存的时候,ECC使用一种特殊的算法对数据进行计算,其结果称为校验位(check bits)。
将所有校验位加在一起的和是校验和(checksum ),校验和与数据一起存放。
当这些数据从内存中读出时,采用同一算法再次计算校验和,并和前面的计算结果相比较,如果结果相同,说明数据是正确的,反之说明有错误,ECC可以从逻辑上分离错误并通知系统。
当只出现单比特错误的时候,ECC 可以把错误改正过来不影响系统运行 (图1).图1: ECC检查并改正单比特错误的原理示意图除了能够检查到并改正单比特错误之外,ECC能检查到(但不改正)单DRAM芯片上发生的任意2个随机错误,并最多可以检查到4比特的错误。
当有多比特错误发生的时候,ECC内存会生成一个不可隐藏(non-maskable interrupt)的中断(NMI),系统会中止运行以避免出现数据恶化。
ECC技术为很多应用软件都提供了有效的保护。
但如图2所示,随着内存容量的增加,ECC的有效性在降低。
(服务器的损耗从3%上升到48%) 有两个因素促使工业标准服务器不断提高其内存的容量:操作系统可以支持更大的内存;低价、高容量的内存越来越普遍。
图2:一年内因为内存故障造成的服务器宕机故障新的ECC内存为了加强对内存数据的保护,惠普于1996年引入了新的ECC技术(Advanced ECC technology)。
目前惠普和其它的服务器制造商的生成线上生产的仍然是带有此类技术的产品。
标准的带有ECC技术的设备在从DIMM中读数据的时候可以同时纠正数据中出现的单比特错误,新的ECC技术可以纠正DRAM中发生的多比特错误,保证DRAM 芯片上的数据全部有效。
在带有新的的4-bit (X4) ECC技术的内存中,每一个芯片“贡献”4个比特的数据给数据字。
从每个芯片中出来的这4比特数据被平均分配给4个带有ECC技术的设备,每个设备一个比特,这样一来,每个芯片上发生的错误可以分解为4个独立的单比特错误。
图 3是这种技术实现的示意图。
图3:在新的ECC技术中每个 DRAM 芯片输出4比特数据分配给4个ECC驱动器。
每一个ECC设备可以纠正单比特错误,改进后的ECC设备可以纠正一个DRAM芯片中发生的多比特错误,因此新的ECC技术为整个设备提供故障保护。
(表1).表1虽然改进后的ECC可以保护数据避免出现错误,但是它只能纠正发生在单个DRAM芯片中的数据错误,而不支持故障恢复和热插拔能力,对于长时间不间断运行的商用软件而言,要实现其软件功能,就要求系统具备这两种能力。
否则,当内存出现了问题的时候,系统只能关机,等待更换新的内存。
最新一代的惠普ProLiant服务器提供3级AMP技术,拥有更强的容错能力,能够满足应用软件对高可靠性的要求。
惠普新的内存保护技术惠普致力于开发内存保护技术,并将这些技术应用到ProLiant系列服务器中,以提高系统的容错性,增加系统的可信度。
(参看图 4)。
惠普ProLiant 300, 500, 和700 系列服务器都具备一项或多项AMP 技术:在线备份内存(Online Spare Memory), 热插拔镜像内存(Hot Plug Mirrored Memory),和热插拔RAID内存(Hot Plug RAID Memory)。
图4 惠普最新一代ProLiant服务器中的高级内存保护技术惠普AMP技术支持工业标准的256-MB, 512-MB, 1-GB DIMM和 2-GB DDR DIMM 。
惠普ProLiant 300系列服务器采用的内存保护技术在线备份内存模式相对标准内存模式而言,在线备份内存模式可以提供更高级别的内存保护。
它特别适用于没有足够计算机系统维护人员或者手头没有多余的内存可以方便的替换的场所。
惠普ProLiant ML370 G2 和DL380 G2 服务器的主板上有6个DIMM插槽。
这些插槽分成3个存储区(memory banks)A、B和C。