This is watching 3 drives being tar-ed | wc -c in parallel (I suspect
a bad drive, trying to narrow it down). The machine is a dual 1.6ghz
Athlon, Tyan motherboard, 2GB of ram.
I'm a little surprised at the system time - yeah, 72MB/sec is a far amount
of I/O but should we really be eating up that much system time?
load free cach swap pgin pgou dk0 dk1 dk2 dk3 ipkt opkt int ctx usr sys idl
3.16 11M 1.4G 0 74M 2.4K 0 0 0 0 0 0 1.3K 46K 0 92 108
3.23 11M 1.4G 0 71M 2.4K 0 0 0 0 0 0 1.3K 45K 1 88 111
3.29 11M 1.4G 0 72M 2.4K 0 0 0 0 0 0 1.2K 46K 2 105 93
3.35 11M 1.4G 0 73M 3.2K 0 0 0 0 0 0 1.2K 45K 1 100 99
3.32 11M 1.4G 0 74M 2.4K 0 0 0 0 0 0 1.2K 45K 1 98 101
3.53 11M 1.4G 0 74M 65K 0 0 0 0 0 0 1.3K 46K 4 103 93
I thought maybe it was the | wc -c (notice the context switches/sec) so I
tried it with a cpio into /dev/null (tar neatly "optimizes" out any I/O
when you are going to /dev/null, I really hate that) and it was still high
load free cach swap pgin pgou dk0 dk1 dk2 dk3 ipkt opkt int ctx usr sys idl
0.78 12M 1.4G 0 66M 34K 0 0 0 0 0 0 945 2.0K 6 82 112
0.96 11M 1.4G 0 66M 2.4K 0 0 0 0 1.0 1.0 964 2.1K 6 79 115
1.12 11M 1.4G 0 66M 2.4K 0 0 0 0 0 0 1.0K 2.3K 7 82 111
1.27 11M 1.4G 0 66M 2.4K 0 0 0 0 0 0 972 2.1K 4 80 116
1.41 11M 1.4G 0 67M 2.4K 0 0 0 0 0 0 960 2.1K 5 87 108
1.62 11M 1.4G 0 67M 2.4K 0 0 0 0 0 0 1.0K 2.3K 7 83 110
--
---
Larry McVoy lm at bitmover.com http://www.bitmover.com/lm