This is not a complaint.
I'm very happy with 400MB/sec through a cheapXXXX power efficient system.
I'm trying to deduplicate some of my 10,000,000+ files.
Top reports spin (mostly on CPU0) up to 10%.
I'm curious which resource is being competed for.
If a waterfall graph would answer my question
any pointers to instructions would be gratefully taken.
Running 3 instances of
find <a disk> -type f -exec cksum {} + > <resultfile>
each searching a different physical drive and resultfile is on
yet another physical drive.
CPU is AMD 5600G w/6 cores & 64MB
top says:
---------
CPU00 states: 15.9% user, 0.0% nice, 5.5% sys, 3.4% spin, 3.0% intr,
72.1% idle
CPU01 states: 16.8% user, 0.0% nice, 4.5% sys, 1.3% spin, 0.0% intr,
77.4% idle
CPU02 states: 17.3% user, 0.0% nice, 5.9% sys, 0.5% spin, 0.0% intr,
76.3% idle
CPU03 states: 11.9% user, 0.0% nice, 4.9% sys, 0.3% spin, 0.0% intr,
82.9% idle
CPU04 states: 9.1% user, 0.0% nice, 2.4% sys, 0.5% spin, 0.0% intr,
88.1% idle
CPU05 states: 5.5% user, 0.0% nice, 1.2% sys, 0.2% spin, 0.0% intr,
93.1% idle
iostat says:
tty sd2 sd3 sd4 cpu
tin tout KB/t t/s MB/s KB/t t/s MB/s KB/t t/s MB/s us
ni sy sp in id
519 89 64.00 2496 156.00 64.00 1991 124.44 64.00 2061 128.81
12 0 4 0 1 83
519 267 64.00 2551 159.44 64.00 2087 130.44 64.00 2142 133.88
13 0 4 1 0 82
Secondary question:
This scenario hits some transfer rate limit.
It's not obviously a chipset/memory system limit.
I think that combination can deliver above 1GB/sec to the SATA controller.
All of the drives can transfer at least 140MB/sec, some as high as 180.
I believe that sd3 and sd4 can transfer faster than sd2.
Is it likely that the kernel services interrupts, etc. in drive # order?
(Reading the Source doesn't give an obvious answer.
There are interactions with scheduling)
thanks
Geoff Steckel
