hp方案2-aac项目vpar竞争分析emc

1、3PAR 竞争分析EMC Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.议程EMC Vmax竞争分析EMC VNX竞争分析2 Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.3PAR vs EMC Vmax

2、 Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.HP 3PAR vs. EMCSS10400SS10800Silver Bullet for EMCSS7400四控SS7400双控SS7200传统技术性能不随容量增长管理复杂，难以变化存在热点盘，性能有限主/从控制器不能双活MiSt- 多控对性能有限控制器损坏性能陡降架构复杂10+种架构，难以管理- 升级，投资难保护VNX2- 7600/8000/

3、VMAX10KVNX2- 5400/5600VNX2- 5600/5800VMAX20KVMAX40K Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.VMAX引擎 = VNX双控 + S-RIO交换机互联并不存在所谓的Global Memory全局共享缓存，物理缓存全部，采用简单的索引在每个主控板和/转发缓冲区进行跨主控板的内存调用采用CMI进行主控板之间的内存镜像，与Clariion/VNX完全相

4、同所有内存完全镜像，包括了控制内存/写内存/读内存使用与VNX相同的DAE，每个DAE两块LCC卡(卡A连odd Dir.，卡B连even Dir.)，多个DAE雏菊链级联Director 控制器VMAX的引擎才是独立控制器，Dir.只是集成了CPU和Memory的主控板整个引擎是一个故障单点，主控板只是这个故障单点的一部分，两个主控板必须通过引擎内的主 动式中板镜像内存、交换信息、共享主机端口卡槽和其它主动式中板 shared前端接口卡槽- shared管理模块- shared风扇 shared电源- sharedVMAX的引擎互联VMAX10K2.0 x8VMAX20K1.0 x8VMAX

5、40K2.0two x8PCIe版本PCIe通道PCIe带宽/主控 8*500MB/s=4 8*250MB/s=2 16*500MB/s=8板PCIe“聚合”带宽/引擎实际的PCIe系统“聚合”带宽 S-RIO版本S-RIO端口通道 S-RIO端口/引擎S-RIO “聚合”带宽/主控板S-RIO端口/引擎S-RIO“聚合”带宽/引擎实际的S-RIO系统“聚合”带宽 EMC宣称的“聚合”带宽/引擎 EMC宣称的系统 “聚合”带宽GB/s8GB/sGB/s4GB/sGB/s16GB/s32GB/s2.0 x4232GB/s2.0 x42128GB/s2.0 x4410GB/s10GB/s20GB/

6、s44820GB/s20GB/s40GB/s80GB/s80GB/s320GB/s50GB/s24GB/s50GB/s200GB/s192GB/s400GB/s业界主流S-RIOVer2.0(6.25Gbaud)，最大聚合带宽交换机带宽30GB/s，12个x4端口(2.5GB/s)，100ns时延VMAX缓存镜像可以在Engine镜像也可以跨Engine镜像(不)奇偶搭配1对1，镜像的“对象”在系统IMPL时决定一个Dir.故障导致的结果是两个Dir.实际上都不可用每个Loop的前5块盘上的部分空间用于沉降，共40 x9GB=360GB，存放物理缓存的两份逻辑拷贝(数据和元数据)，SPS继续供

7、电5分钟最大472GB Global Memory（跨Dir.镜像后）实际数据缓存在35%60%，取决于各种Data Servi（TimeFinder、SRDstVP等）的使用情况，取50%作为均值，假定配置512GB缓存，实际可用的读写缓存在镜像后只有128GB，再扣除SFB的部分（假设占1/4），实际真实读写缓存只有96GB为每个Volume分配固定数量的cache slot(在高负载时可以微量调整，最大是基数的三倍)，而不能在需要时将所有 cache用于单个卷3PAR缓存容量vs. VMAXVMAX10K512128646480r/20w NA3212860r/40w NA3212810

8、400384963264VMAX40K204825612812810800768963264系统缓存容量(GB)单个独立控制器缓存容量(GB)控制缓存(GB)数据缓存(GB)可用读缓存(GB)可用写缓存(GB)镜像后实际可用数据缓存容量(GB)系统实际可用数据缓存容量(GB)42.610.753.3213.242.610.753.3426.4NA64512可用读缓存(GB)可用写缓存(GB)镜像后实际可用数据缓存容量(GB)系统实际可用数据缓存容量(GB)27.418.345.7182.827.418.345.7365.6NA64512VMAX后端连接非Daisy Chain级联的DAE只有3

9、2/64个使用低密度DAE(2.5&3.5)=480/960块盘使用高密度DAE(2.5)=800/1600块盘VMAX的性能饱和点和VNX类似只有标称容量的一半！VMAX的后端连接不均衡使用低密度DAE时实际最大磁盘数量是960/1920而不是2400使用高密度DAE时实际最大磁盘数量是1560/3200VMAX20K/40K的低配严重浪费机柜空间后端连接使用HSSDC铜缆，机柜间距只得25米增加Engine？要均衡！VMAX的后端连接不均衡！VMAX BP：不要分区要均衡！尤其不要Engine分区！What can Happen with SegregationWhat can Happe

10、n with SegregationWhat can Happen with Segregation3PAR vs. VMAX处理能力VMAX10K104002x4core 2.8GHz 2xHarrier 8x4core 2.8GHz 8xHarrier32VMAX40K4x6core 2.8GHz 032x6core 2.8GHz064108002x4core2.8GHz2xHarrier 16x4core 2.8GHz16xHarrier64单个独立控制器核数2x6core 2.8GHz 08x6core 2.8GHz016系统核数系统前端核数（1/3） 系统后端核数（2/3）PCIe单

11、独控后端口系统后端口322.04 Mod (4 QSFP)64x4Gb FC8xHarrier2.012x8Gb FC48x8Gb FC1282.04 Mod (4 QSFP)128x4Gb FC16xHarrier2.012x8Gb FC96x8Gb FC的前端口！VMAX的cores被逻辑分区后绑定在特定的前后端口上，不能充分共享和完全利用RT很重要？50%！超过50%？准备升级吧！陈旧的VMAX VP(1)ck = single thin device extent = 12 tracks = 768KB FBA emulation在回收时，整个ck被装入cache检查是否连续的0空间，

12、如果全0则TimeFinder卷不可回收(2)extent group =10 contiguous thin device extents = 120 tracks = 7.5MB FBA(3)region = extent group set = 48 contiguous extent groups = 5760 tracks = 360MB FBAcapacity，default to 10%(4)PRC = pool(5)one cache slot for each thin device = 1 track = 64KB (6)T10 UNMAP不支持：预留空间ORS占用设备RP

13、占用设备FTS设备ap目标设备和TF/VP Snap目标设备TRAID SelectionVP General BPVP BP：Striped Meta3PAR使精简成为高端的标配！PlatformMeasured IOPS degradation during Zero Page ReclaimMeasured Host Latency Degradation during Zero Page ReclaimEMC VMAX48.19%40.58%Dell Compellent42.33%74.22%EMC VNX29.99%42.61%HDS VSP23.34%30.33%HP 3PAR0

14、%1.23%Thin ProviOut of the EMC Whiteping with the EMC VMAXr ” EMC Symmetrix VMAX Virtual Proviing SpaceReclamation and Application Considerations”:Starting the space reclamation pros will spawn a back-end disk director (DA) tasktwill examine the allocated thin device extents on specified thin devi.

15、A thin device extentis 768 KB (or 12 tracks) in size and is the default unit of storage at which data is allocated for a thin device. For each allocated extent, all 12 tracks will be broughto Symmetrix cache and examined to see if they contain all zero data. If the entire extent contains all zero da

16、ta, the extent will be de-allocated and added backo the pool, making it available for a new extent allocation operation.There is no practical way of knowing how long a vmkfstools -y operation will take tocomplete. It can run for a few minutes or even for a few hours depending on the size of thedatas

17、tore, the amount of dead space and the available resouron the array to executethe operation. Reclamation on the Symmetrix is a low-priority task (in order for it to notnegatively affect host I/O performance) so a heavy workload on the device1, pool or the array itself can throttle the pros and slow

18、the reclamation.Thin ProviOut of the EMC Whiteping with EMC VNXr ” VIRTUAL PROVIING FOR THE NEW VNX SERIES”When to use Classic, Thick, and ThinsIt is important to understand your application requirements and select the approacht meets your needs. Ifconditions change, you can use VNXmigration to migr

19、ate among thin, thick, and classics.Use pool-based thins for:Applications with moderate performance requirementsTaking advantage of advanced data servilike FAST VP, VNX Snapshots, Compresand Deduplication;Ease of setup and management; Best storage efficiency; Energy and capital savings; Applications

20、 where spaceconsumption is difficult to forecastUse pool-based thicks for:Applicationst require good performanceTaking advantage of advanced data servilike FAST VP and VNX Snapshots; Storage assigned to VNX for File; Ease of setup and managementUse classics for:Applicationst require extreme performa

21、nce (for exle, when milliseconds of performance are critical)The most predictable performance; Precise data placement on physical drives and logical data objects; Physicalseparation of data3PAR vs EMC VNX Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibjec

22、t to change without notice.EMC VNX M Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX Series Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.架构EM

23、C VNX Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX 硬件架构 Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX 硬件架构 Copyright 2012 Hewlett-P

24、ackard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX MCx Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX MCx Architecture Copyright 2012 Hewlett-Packard Development

25、Company, L.P. The information contained herein ibject to change without notice.EMC VNX Cache Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX RAID Copyright 2012 Hewlett-Packard Development Company, L.P. The information

26、 contained herein ibject to change without notice.EMC VNX Software Suite Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX Thin Provi Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained her

27、ein ibject to change without notice.EMC VNX Pool Architecture Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.EMC VNX FAST VP Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein ibject to change without notice.Thin ProviOut of the EMC Whiteping with EMC VNXr ” VIRTUAL PROVIING FOR THE NEW VNX SERIES”When to use Classic, Thick, and ThinsIt is important to understand your application requirements and select the approacht meets your needs. Ifconditions change, you can use VNXmig