HDS及服务器维护大7).ppt

《HDS及服务器维护大7).ppt》由会员上传分享，免费在线阅读，更多相关内容在PPT专区-天天文库。

1、主题从磁盘谈起AMS控制器的后端架构CacheSegmentSize与LUNStripeSizeLunOwnerShip与DataShareMode主机设置的建议TheHardwareThePlatterMadeofglassoraluminumCoatedwithmagneticoxideFormattedintosectorsandtracksRead/WriteHead&CylinderLatency:whenheadmovesalongthetrackonasurfaceasplatterspinSeek:whenheadmovesfromonetracktoanother

2、onthesamesurfaceHeadswitch/skew:whenheadmovesfromonetracktoanotheronthesamecylinderHDDSpecification(MM)USPand9900VarenotthesameIOPSIOPS=1secor1000ms_(seektime+averagelatency)MeaningfulonlyforrandomsmallblocksizeI/O,upto8KBblockAvg14615KrpmreadIOPS=1000ms/(3.8+2.01)=172Avg14615KrpmwriteIOPS=100

3、0ms/(4.1+2.01)=164Avg30010KrpmreadIOPS=1000ms/(4.9+2.99)=127Avg30015KrpmwriteIOPS=1000ms/(5.5+2.99)=118Writeseektakeslongertimebecauseitdemandsforbetterprecisionintrackpositioning15Krpmisbetterthan10KrpmRPMisthefactor,notsizeofHDD1x146G=2x72Gincapacity,NOTIOPSAMS500控制器架构AMS500后端架构AMS500后端环路Que

4、stion:一个RAIDGROUP的硬盘来自与不同的扩展柜，对性能有提升吗？AMS后端架构IOPSIOPS=1secor1000ms_(seektime+averagelatency)MeaningfulonlyforrandomsmallblocksizeI/O,upto8KBblockAvg14615KrpmreadIOPS=1000ms/(3.8+2.01)=172Avg14615KrpmwriteIOPS=1000ms/(4.1+2.01)=164Avg30010KrpmreadIOPS=1000ms/(4.9+2.99)=127Avg30010KrpmJwriteIOPS

5、=1000ms/(5.5+2.99)=118Writeseektakeslongertimebecauseitdemandsforbetterprecisionintrackpositioning15Krpmisbetterthan10KrpmRPMisthefactor,notsizeofHDD1x146G=2x72Gincapacity,NOTIOPSHence,roughlywesaideachHDDprovide150IOPSRaidLevelRAIDGROUP的建议RAID1+0从2D+2D~4D+4DRAID5从3D+1P~8D+1PRAID6适用于数据的安全性要求很高

6、，同时是顺序读写的环境CacheSegmentSize与LUNStripeSizeCacheSegmentSize与LUNStripeSizeCacheSegmentSize默认16KCacheSegmentSize可通过建立不同的CachePartition，定义不同的SegmentSize:4K8K16K256K512KLUNStripeSize默认64K可根据需要调整为256K512KOLTP的情况下，建议采用默认的CacheSegmentSize16KLUNStripeSize64K中端盘阵里，LUN有Ownership这个概念,在同一时刻，一个LUN只能被一个控制器接管D

7、ataShareModeDataShareModeDataShareMode当使用HDS的多路径软件HDLM的时候DataShareMode不是必须的当使用第三方的多路径软件，而该软件是不能识别AMS的LUNOwerShip，同时该软件是走LoadBalance的，建议设置此项问题使用VertiasDMP需要吗?对主机的建议LUN的分配与主机参数的建议尽量不要把一个应用全压到一个RAIDGROUP上，尽可能地分布在不同的RAIDGROUP上尽量避免顺序IO的应用与随机