惠普HP大容量網(wǎng)絡(luò)存儲(chǔ)解決方案培訓(xùn)

1、惠普大容量存儲(chǔ)解決方案 2008 hewlett-packard development company, l.p.本文所含信息如有更改，恕不另行通知。8000pb500040003000存儲(chǔ)的未來w w disk storage systems petabytes, 2006-2010 forecastpb18000存儲(chǔ)消耗加速增長(zhǎng)160001400020082010年的增長(zhǎng)速度將超過50%12000100006000400020000may 2006 forecast2006200720082009201090008000w w file and block disk storage sy

2、stems, 2005-2008資料來源： 2006年idc調(diào)70006000block basedfile based向基于文件的存儲(chǔ)轉(zhuǎn)變2008年，用戶需要使用專門存儲(chǔ)文件數(shù)據(jù)的磁盤存儲(chǔ)塊數(shù)據(jù)2000100002005200620072008資料來源： 2007年idc調(diào)查報(bào)告當(dāng)前的業(yè)務(wù)模式需要新型基礎(chǔ)設(shè)施在線社區(qū)照片共享與潤色服務(wù)借助技術(shù)應(yīng)用的全球化以及速度更快、性能更高、價(jià)格更低的技術(shù)來實(shí)現(xiàn)流媒體視頻服務(wù)音樂與鈴聲服務(wù)內(nèi)容共享數(shù)字動(dòng)畫醫(yī)療影像分析安全監(jiān)控系統(tǒng)基因組研究能源勘探問題寬帶服務(wù)手機(jī)廣播生命科學(xué)石油天然氣企業(yè)應(yīng)用程序、相機(jī)和手機(jī)的文件容量越來越大越來越多的用戶需要觀看和共享視頻

3、廣播和社區(qū)網(wǎng)站的用戶數(shù)量呈指數(shù)級(jí)增長(zhǎng)存儲(chǔ)消耗正在以前所未有的速度增長(zhǎng)2008年11月14日6星期五實(shí)際環(huán)境中的大容量存儲(chǔ)問題照片共享服務(wù)提供商在線訂購用戶的數(shù)量呈指數(shù)級(jí)增長(zhǎng)圖像捕獲技術(shù)飛速發(fā)展周期性用戶活動(dòng)流媒體技術(shù)提供商問題用戶數(shù)量隨時(shí)都會(huì)激增媒體技術(shù)飛速發(fā)展業(yè)務(wù)需求瞬息萬變電信服務(wù)提供商機(jī)遇當(dāng)前的基礎(chǔ)設(shè)施已無法滿足需求值得信賴且擁有用戶基礎(chǔ)的合作伙伴不斷增加的收入來源惠普機(jī)密。僅供惠普及其渠道合作伙伴內(nèi)部使用。etata6previous solutions have bottleneckscentralized architecturesmetadata servermdadatadat

4、anamespace serverlock managertoken managerdoesnt scale most enterprise applicationworkloads are metadata intensivesmall files, small i/os, randomaccess, metadata operationscentral component is a bottleneckdistributed lock managersany server can modify anyportion of the file systemdoesnt scaleinter-s

5、erver messaging grows as(number of servers)2messaging overhead too highbeyond few (10) serverscomplex failure modes2 april 2010hp storageworks x9720extreme data storage systemintegrated hardware/software solution1. software proven in content rich environments integrated system management simplified

6、performance and capacity scaling multiple access protocols nfs, cifs, x90002. performance blocks scale performance up to 12.8 cores/u starting at 3 blades, expand up to 16 blades quad-core, half-height blades in c7000 enclosure3. capacity blocks industry leading density up to 12tb/u starting at 82tb

7、, expand to 656tb scales in 82tb capacity blockshp restricted. hp and channel partner internal use.msa2000msa2000msa20008x9320 capacity/performancecapacity couplet disk assignmentsperformance couplet disk assignmentssata sparevdisk 1vdisk 1sata sparevdisk 2vdisk 2sata sparevdisk 3vdisk 3sata sparevd

8、isk 4vdisk 4vdisk 1vdisk 1vdisk 1vdisk 2vdisk 2vdisk 2vdisk 3vdisk 3vdisk 3vdisk 4vdisk 5vdisk 5vdisk 1vdisk 1vdisk 1vdisk 2vdisk 2vdisk 3vdisk 4vdisk 4vdisk 4vdisk 5vdisk 5vdisk 5vdisk 1vdisk 2vdisk 2vdisk 3vdisk 3vdisk 3vdisk 4vdisk 4vdisk 4vdisk 5vdisk 5vdisk 5sas sparevdisk 1vdisk 1vdisk 3vdisk

9、3vdisk 3sas sparevdisk 5vdisk 5vdisk 7vdisk 7vdisk 7vdisk 1vdisk 1vdisk 1vdisk 3vdisk 3vdisk 3vdisk 5vdisk 5vdisk 5vdisk 7vdisk 7vdisk 7vdisk 2vdisk 2vdisk 2vdisk 4vdisk 4vdisk 4vdisk 6vdisk 6vdisk 6vdisk 8vdisk 8vdisk 8vdisk 2vdisk 2vdisk 2vdisk 4vdisk 4vdisk 4vdisk 6vdisk 6vdisk 6vdisk 8vdisk 8vdi

10、sk 84 global hot spares4x 7+2 and 1x 6+2 raid6 vdisksvdisk1, vdisk3 and vdisk5 assigned to controller avdisk2 and vdisk4 assigned to controller b1 lun per vdisk5 luns *(segments) per couplet2 april 20102 global hot spares6x 5+1 raid5 and 2x 4+1 raid5 vdisksvdisks 1, 3, 5, 7 assigned to controller av

11、disks 2, 4, 6, 8 assigned to controller b1 lun per vdisk8 luns per coupletmsa2000capacity couplet storage configurationno snap reserveibrix file system 31tb usablesegment server 1segment server 2ibrix lvsegment6.5 tb(ext3)ibrix vg6.9 tb lunvdisk11tb sata drivesraid 69 drives (7+2)controller aibrix l

12、vsegment6.5 tb(ext3)ibrix vg6.9 tb lunvdisk21tb sata drivesraid 69 drives (7+2)controller bibrix lvsegment5.4 tb(ext3)ibrix vg5.9 tb lunvdisk31tb sata drivesraid 68 drives (6+2)controller aibrix lvsegment6.5 tb(ext3)ibrix vg6.9 tb lunvdisk41tb sata drivesraid 69 drives (7+2)controller aibrix lvsegme

13、nt6.5 tb(ext3)ibrix vg6.9 tb lunvdisk51tb sata drivesraid 69 drives (7+2)controller bcontroller acontroller bnote: if snapshots are desired, space must be reserved in the luns for snap pools prior to formatting the ext3volsmsa2000msa2000performance couplet storageconfiguration no snap reserveibrix f

14、ile system 15tb usablesegment server 1segment server 2ibrix lvsegment1.5tb(ext3)ibrix vg1.8 tb lunvdisk1 450gbsas hdds raid 5 5 drives(4+1) controller aibrix lvsegment2 tb(ext3)ibrix vg2.25 tb lunvdisk2 450gbsas hdds raid 5 5 drives(5+1) controller bibrix lvsegment2 tb(ext3)ibrix vg2.25 tb lunvdisk3

15、 450gbsas hdds raid 5 5 drives(5+1) controller aibrix lvsegment2 tb(ext3)ibrix vg2.25 tb lunvdisk4 450gbsas hdds raid 5 5 drives(5+1) controller bibrix lvsegment1.5tb(ext3)ibrix vg1.8 tb lunvdisk1 450gbsas hdds raid 5 5 drives(4+1) controller aibrix lvsegment2 tb(ext3)ibrix vg2.25 tb lunvdisk2 450gb

16、sas hdds raid 5 5 drives(5+1) controller bibrix lvsegment2 tb(ext3)ibrix vg2.25 tb lunvdisk3 450gbsas hdds raid 5 5 drives(5+1) controller aibrix lvsegment2 tb(ext3)ibrix vg2.25 tb lunvdisk4 450gbsas hdds raid 5 5 drives(5+1) controller bcontroller acontroller bcontroller acontroller bnote: if snaps

17、hots are desired, space must be reserved in the luns for snap pools prior to formatting the ext3volsisilonx9000 network storage systemcompetitioncompeting productsnetappontap 7g, ontap 8 (gx)ibmgpfs (sofs)competitoradvantages簡(jiǎn)單管理, 市場(chǎng)認(rèn)可, 合作伙伴眾多nas行業(yè)領(lǐng)跑者brand recognition,scalable file system,feature se

18、tcompetitordisadvantages只在特殊領(lǐng)域，cover范圍不廣貴hpc行業(yè)沒有案例不適合scale out 領(lǐng)域復(fù)雜ibm支持差san架構(gòu)112 april 201011hp product announcement hp restrictedtechnical concepthp restricted2009 hewlett-packard development company, l.p.the information contained herein is subject to change without notice.ibrix research and devel

19、opmentinnovation in data distribution and backuprecently awardedu.s. patents*patent #7,464,125checking the validity of blocks andbackup duplicates of blocks duringblock readspatent #7,430,570shadow directory structure in adistributed segmented file systemibrix fusionsegmented file systempatent #7,40

20、6,484storage allocation in a distributedsegmented file system*partial list132 april 2010the ibrix distribution engine (ide)standard os stackos stack w/ideclientsclientsnetwork driversnetwork driverstcp / ipnfs / cifsvfsapi layerpfstcp / ipnfs / cifsvfsapi layeribrixdistribution engineapi layerdevice

21、 driversibrix didnt reinvent thephysical file system.ext3device driversstoragestorage142 april 201014f1f2f3fns1s2s3sns2f1f2f3s1snfns3non-hierarchical architecturedirus patent # 6,782,389f1f2f3fns1s2subdirs3ssn1dirs2extremely highaggregateperformancefrom a singledirectory(and single file)1s2sn423100f

22、1f2f3fns1subdirsns3segmentserverssegments152 april 2010allocation policycontrols placement of new files random choose segment at random roundrobin cycle through permutation of segments directory place file on same segment as its parent directory none (for directories only) use same policy for files

23、and directories automatic choose a segment with the most free space sticky use only one segmentthis policy is generally not recommended. keep in mind that other policies willautomatically redirect file creation and append operations away from segments thatbecome full. preferred segments set of segme

24、nts to place files on first if possible. local for segment server, choose a segment it owns162 april 2010the ibrix segmented concepttraditional file system design: monolithic metadata managementblock 0block n-1obsoleteboundaries (e.g.cylinder groups)segmented file system “distributed file space” app

25、roacheach segment is almost a “mini” file system with its own set of metadataallocation is generally localized within a segment, but may span segmentsdeterministically distributedeach segment owned by exactly one ibrix enginesegmentowner mapping determined locally by each ibrix engineflexible segmen

26、t control can migrate to balance load172 april 2010distributed inode spaceeach segment is a range of inodes client can unilaterally determine two things by looking at the inode numberof an entity: which segment/server owns that entity what is the coordinate of that entity within that segmentinode 0i

27、node 0inode 999 inode 1000inode 999 inode 0inode 1999 inode 2000inode 999 inode 0inode 2999inode 999inode 9000inode 0inode 9999inode 999file1 i_1500file2 i_2200182 april 2010storage topology bricks potential data pathsibrixclient 1ibrixclient 2nfsclient 1nfsclient 2segmentserver 1segmentserver 2segm

28、entserver m-1segmentserver mownssegmentspspspspa13b24filea57b68storage topology bricks ibrix client readibrixclient 1ibrix fileclient 2nfsclient 1nfsclient 2segmentserver 1spasegment fileserver 2spbsegmentserver m-1spasegmentserver mspb4filestorage topology bricks multipleibrix clients read same fil

29、eibrixclient 1fileibrixclient 2filenfsclient 1nfsclient 2segmentserver 1spasegment fileserver 2spbsegmentserver m-1spasegmentserver mspb4filestorage topology bricks ibrixclient 1ibrixclient 2nfs client readfilenfsclient 1nfsclient 2segmentserver 1spasegment fileserver 2spbfilesegmentserver m-1spaseg

30、mentserver mspb4filestorage topology bricks nfs multiple-client readfilefilefileibrixclient 1ibrixclient 2nfsclient 1nfsclient 2nfsclient 2user networkcluster networksegmentserver 1spasegment fileserver 2spb4filefilesegmentserver m-1spafilesegmentserver mspbstorage topology bricks nfs cached readsfi

31、lefilefileibrixclient 1ibrixclient 2nfsclient 1nfsclient 2nfsclient 2segmentserver 1spsegment fileserver 2spfilesegmentserver m-1spfilesegmentserver mspabab4filestorage topology bricks nfs san readsfilefilefileibrixclient 1ibrixclient 2nfsclient 1nfsclient 2nfsclient 2segmentserver 1segmentserver 2filesegment