Hello,
We're following the paper "Performance of Large Journaling File Systems" http://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101148 to evaluate the performance of disk I/O using FCP-SCSI devices. We have two host FCP chpids (X and Y), connected to two storage ports (P and Q). X is connected to P and Y to Q. We work with 4 luns (A, B, C, D) of 100 GB. We reach it with two paths (X-P and Y-Q), so we have 8 paired devices. We are using dm-multipath: cgefls01:~ # multipath -ll mpath0 (360060480000290100625533030463142) dm-0 EMC,SYMMETRIX [size=102G][features=0][hwhandler=0] \_ round-robin 0 [prio=2][active] \_ 0:0:0:115 sda 8:0 [active][ready] \_ 1:0:0:115 sdb 8:16 [active][ready] We ran several iozone benchmark with different options: ./iozone/iozone -i 0 -i 2 -s 2000m -t 16 -r 64k -C -e -w -F {list of files} - with 8 threads (1000m and 2000m file size) - with 16 threads (1000m and 2000m file size) Using 16 threads and 2000m file size, the linux hangs completely. I can't connect to it with ssh or by 3270 (so, no tcp/ip problem). In perfsvm I see it takes 100% CPU, but most of the cpu consumption is for the supervisor. Detailed data for user CGEFLS01 Total CPU : 97.6% Storage def. : 1024MB Page fault rate: .0/s Superv. CPU : 64.3% Resident <2GB: 104303 Page read rate : .0/s Emulat. CPU : 33.3% Resident >2GB: 156848 Page write rate: .0/s I haven't applied the VM64306 apar, but I've tried the NOQIOASSIST workaround with no result (and we have 1GB of guest storage, so it's no related). Environment: - z/VM 5.3 with RSU-0702 - Linux SLES SP1 (up to date), 1GB storage. Root and swap on FICON/ECKD. Test partition on FCP-SCSI. - Two FCP chpids - Benchmarking tools: iozone - EMC DMX storage array and switch Thanks! ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
