netappfc存储性能测试报告.doc
Netapp fc存储性能测试报告目录Netapp fc存储性能测试报告11.-bs = 4k ， -iodepth = 1024，改变线程数量numjobs顺序读模式测试：12.-bs = 4k ， -iodepth = 1024，改变线程数量numjobs，顺序写模式测试：23.-bs = 4k ， -iodepth = 1024，改变线程数量numjobs，同时测试iscsi和fc的顺序写模式。34.-bs = 4k ， -iodepth = 1024，改变线程数量numjobs，混合读写模式，读占70%，写30%，测试：41. -bs = 4k ， -iodepth = 1024，改变线程数量numjobs顺序读模式测试：fio.conf:globalioengine=libaiodirect=1norandommap=1randrepeat=0runtime=10ramp_time=6size=90Gfilename=/dev/sdfiodepth=1024bs=4kreadstonewallgroup_reportingrw=readnumjobs=1readstonewallgroup_reportingrw=readnumjobs=10省略.结果：顺序读numjobsbsiodepthIOPS平均延时msec吞吐率 MB/s14k10248025312.75321.4104k102439208026.081535.6204k102439470351.761549.9304k102439605977.261559.6404k1024396328102.931564.9504k1024397119128.231571.2604k1024400640152.41588.1704k1024400486177.411592.2804k1024400675202.671596.1904k1024396721229.761585.41004k1024396463254.981588.4结论：-bs = 4k ， -iodepth = 1024固定时顺序读，线程数量到达60后， iops、吞吐率趋于稳定，总延时持续增长。iops最大400675，吞吐率最大1592.2MB/s2. -bs = 4k ， -iodepth = 1024，改变线程数量numjobs，顺序写模式测试：fio.conf:globalioengine=libaiodirect=1norandommap=1randrepeat=0runtime=10ramp_time=6size=90Gfilename=/dev/sdfiodepth=1024bs=4kwritestonewallgroup_reportingrw=writenumjobs=1writestonewallgroup_reportingrw=writenumjobs=10省略.结果：writenumjobsbsiodepthIOPS平均延时msec吞吐率 MB/s14k10241710765.98684.7104k102432483631.51272.1204k102432483631.51298.3304k102433025361.781298.3404k1024321176126.921270.6504k1024304834166.561210.6604k1024283171214.811129.2704k1024279707255.911120.5804k1024291367277.651169.1904k1024289363313.561165.11004k1024272198371.631102.7结论：-bs = 4k ， -iodepth = 1024固定时顺序写，线程数量到达30后， iops、吞吐率到达最高，30后开场下降，总延时持续增长。iops最大330253，吞吐率最大1298.3MB/s。3. -bs = 4k ， -iodepth = 1024，改变线程数量numjobs，同时测试iscsi和fc的顺序写模式。Fio.conf指令同测试2，另外写一个fio_iscsi.conf。同时运行fio fio.conf 和fio fio_iscsi.confFio_iscsi.conf:globalioengine=libaiodirect=1norandommap=1randrepeat=0runtime=10ramp_time=6size=90Gfilename=/dev/sdjiodepth=1024bs=4kwritestonewallgroup_reportingrw=writenumjobs=1writestonewallgroup_reportingrw=writenumjobs=10省略.iscsi和fc同时fio测试write时，fcnumjobsbsiodepthIOPS平均延时msec吞吐率 MB/s14k10241170598.74468.649104k102426855838.081053.6204k102428693871.071128.9304k1024285463107.081127.6404k1024289235140.641145.8504k1024277900182.561105.5604k1024274672220.891096.8704k1024269146264.641079.2804k1024270521300.671088.5结论：-bs = 4k ， -iodepth = 1024固定时顺序写，iscsi的写操作对fc的写操作有影响，iops和吞吐率都有减少。线程数量到达40后， iops、吞吐率到达最高，40后开场下降，总延时持续增长。iops最大289235，吞吐率最大1145.8MB/s。4. -bs = 4k ， -iodepth = 1024，改变线程数量numjobs，混合读写模式，读占70%，写30%，测试：fio_mi*_rw_fc.conf：globalioengine=libaiodirect=1norandommap=1randrepeat=0runtime=10ramp_time=6size=90Gfilename=/dev/sdfiodepth=1024bs=4kwritestonewallgroup_reportingrw=randrwrwmi*read=70numjobs=1writestonewallgroup_reportingrw=randrwrwmi*read=70numjobs=10省略.结果：读写混合，读占70%readwritenumjobsbsiodepthIOPS平均延时msec吞吐率 MB/sIOPS平均延时msec吞吐率 MB/s14k10243292022131.971407921.656.44104k102442037169.75171.00817949169.0873.028204k102439070363.56161.99716806363.1169.676304k102439541529.05166.73416931528.4371.41404k102436802749.43158.61315700748.1867.744504k102434730973.03153.24114903971.4265.648604k1024383651076.79170.613164361075.4173.109704k1024387211214.85174.904166421213.5175.155804k1024382061411.52175.516163501411.8875.224结论：使用混合读写模式，读和写的吞吐、iops都比顺序读、写模式下降了很多，读io的最高吞吐率是顺序读的近十分之一。混合读写模式对磁盘io影响很大，原因尚在查。另外我测了-rw=randrw -rwmi*read=100 ，看起来读占了100%，应该等同于-rw=randread随机读模式，但是结果不同相差很大。两种指令：指令结果fio -filename=/dev/sdf -direct=1 -iodepth 1024 -thread -rw=randrw -rwmi*read=100 -ioengine=libaio -bs=1024k -size=50G -numjobs=30 -runtime=10 -group_reporting -name=mytestbw=391200KB/s, iops=382，lat avg=4969.31fio -filename=/dev/sdf -direct=1 -iodepth 1024 -thread -rw=randread -ioengine=libaio -bs=1024k -size=50G -numjobs=30 -runtime=10 -group_reporting -name=mytestbw=699903KB/s, iops=683, lat avg=43743.14
