David and Vasu, >>agree with vasu - assuming you're trying to hit numbers a vendor spec'd however on a flash drive of >>some sort. >>you're doing random writes which is also tough on some flash drive if they're into write >>amplification. try changing to random reads and you only need a couple of threads (with a depth of >>128 each) to generate enough iodepth to saturate most drives. 64 threads might be more threads >>than you have CPUs to run them so you're switching which also hurts performance I test against a SAN LUN (Infinidat F2000). This my AIX LUN: # lsattr -El hdisk6 PCM PCM/friend/NFINIDATpcm Path Control Module False PR_key_value Persistent Reserve Key value True algorithm round_robin Algorithm True hcheck_cmd test_unit_rdy Health Check Command True hcheck_interval 60 Health Check Interval True hcheck_mode nonactive Health Check Mode True lun_id 0x1000000000000 Logical Unit Number ID False lun_reset_spt yes LUN Reset Supported True max_transfer 0x80000 Maximum TRANSFER Size True node_name 0x5742b0f00004c700 FC Node Name False pvid none Physical volume identifier False q_type simple Queuing TYPE True queue_depth 64 Queue DEPTH True reserve_policy no_reserve Reserve Policy True rw_timeout 30 READ/WRITE time out value True scsi_id 0x1e0800 SCSI ID False unique_id 3B1F742b0f0000004c700000000000001d609InfiniBox08NFINIDATfcp Unique device identifier False ww_name 0x5742b0f00004c711 FC World Wide Name False # With this conf: # cat randwrite.fio [global] thread numjobs=1 iodepth=1 group_reporting bs=4k norandommap=1 refill_buffers direct=1 ioengine=posixaio runtime=300 time_based filename=/dev/hdisk6 log_avg_msec=1000 [randread_32_128] rw=randread numjobs=32 iodepth=128 stonewall # I have now bad resulsts: IOPS: 2 800 Bandwith:11 MB/s latency : 6ms before with old configuration file: IOPS: 42 000 Bandwith:164 MB/s latency : 0.2 ms Regards, Thierry-- To unsubscribe from this list: send the line "unsubscribe fio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
