Hello,
I'm trying to benchmark random reads and I see some strange happenings
on the server.
The idea was to have 128 processes randomly reading from a 10GB file and
log bandwith and latencies. Here's my job file:
---------------------------
[global]
rw=randread
size=10g ; filesize = 10GB
bs=4k ; use block size of 4kB
numjobs=16 ; 16 threads per job
ioengine=libaio ; use async IO
directory=/data ; use /data for testing
iodepth=32 ; queue depth
direct=1 ; to queue up multiple IOs
runtime=3600 ; run for 1 hour
write_bw_log=/home/mikem/fio-logs ; gather bandwidth info
write_lat_log=/home/mikem/fio-logs ; gather latency info
disk_util=1 ; generate disk useage stats
; if supported on the platform
gtod_cpu=10 ; get time of day on cpu 10
[job1] ; 10 jobs all doing the same thing as
; desrcibed in global section
[job2]
[job3]
[job4]
[job5]
[job6]
[job7]
[job8]
---------------------------------------
What I got instead was 128 files of 10GB each with fio 128 processes
running. After the writing of all the files I see
fio: terminating on signal 15
constantly scrolling in the window in which I started the fio job.
An strace -f -p pid shows:
kill(pid, SIGTERM) = 0
If I run `iostat -x /dev/sda1 5 5` the first line of output shows maybe
400 MB/sec being read. Sometimes it spikes up to 800 MB/sec. Now it's
showing 176 MB/sec being written. All subsequent lines show zero across
the board. That's only with the writing phase. The reading phase does
not exhibit that behavior. So I'm not sure if it's doing anything good.
Can someone tell me what I'm doing wrong? Again, my goal is to benchmark
some number of processes or threads randomly reading from some large file.
Thanks,
mikem
--
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
