PostgreSQL的磁盘空间占用问题

　　Remember,thisisadatabaseclusterwidesetting.Sobumpitupenoughtocoveralldatabasesinyourdatabasecluster.Also,eachFSMpageuses6bytesofRAMforadministrativeoverhead,soincreasingFSMsubstantiallyonsystemslowonRAMmaybecounter-productive.　

　　3.4max_fsm_relations

　　Thissettingdictateshowmanynumberofrelations(tables)willbetrackedinfreespacemap.Againthisisadatabasecluster-widesetting,sosetitaccordingly.Inversion7.3.3andlater,thisparametershouldbesetcorrectlyasadefault.Inolderversions,bumpitupto300-1000.　

　　3.5wal_buffers

　　ThissettingdecidesthenumberofbuffersWAL(WriteaheadLog)canhave.Ifyourdatabasehasmanywritetransactions,settingthisvaluebithigherthandefaultcouldresultbetterusageofdiskspace.Experimentanddecide.Agoodstartwouldbearound32-64correspondingto256-512Kmemory.　

　　4Othertips

　　4.1Checkyourfilesystem

　　OnOSlikeLinux,whichoffersmultiplefilesystems,oneshouldbecarefulaboutchoosingtherightonefromaperformancepointofview.ThereisnoagreementbetweenPostgreSQLusersaboutwhichoneisbest.　

　　Contrarytopopularbelief,today'sjournalingfilesystemsarenotnecessarilyslowercomparedtonon-journalingones.Ext2canbefasteronsomesetupsbuttherecoveryissuesgenerallymakeitsuseprohibitive.DifferentpeoplehavereportedwidelydifferentexperienceswiththespeedofExt3,ReiserFS,andXFS;quitepossiblythiskindofbenchmarkdependsonacombinationoffilesystem,disk/arrayconfiguration,OSversion,anddatabasetablesizeanddistribution.Assuch,youmaybebetteroffstickingwiththefilesystembestsupportedbyyourdistribution,suchasReiserFSforSuSELinuxorExt3forRedHatLinux,nottoforgetXFSknownforit'slargefilesupport.Ofcourse,ifyouhavetimetoruncomprehensivebenchmarks,wewouldbeinterestedinseeingtheresults!