If I run " dd if=/dev/zero bs=1024k of=file count=1000 " iostat shows me:
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 671.50 88.00 113496.00 176 226992
However postgres 8.3.7 doing a bulk data write (a slony slave, doing
inserts and updates) doesn't go nearly as fast:
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 418.41 648.76 7052.74 1304 14176
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 237.50 44.00 3668.00 88 7336
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 221.50 444.00 3832.00 888 7664
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 166.00 248.00 3360.00 496 6720
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 163.00 480.00 3184.00 960 6368
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 102.50 724.00 1736.00 1448 3472
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 295.50 712.00 6004.00 1424 12008
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 109.45 433.83 2260.70 872 4544
Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn
sda 180.00 640.00 3512.00 1280 7024
top shows the cpu usage of the pg process ranges from zero to never more
than ten percent of a cpu, and that one cpu is always ninety some odd
percent in iowait. So what is postgres doing (with fsync off) that
causes the cpu to spend so much time in iowait?
This is a 64 bit amd linux system with ext3 filesystem. free shows:
total used free shared buffers cached
Mem: 8116992 8085848 31144 0 103016 3098568
-/+ buffers/cache: 4884264 3232728
Swap: 6697296 2035508 4661788