DragonFly-2.1.1.436.g03a48 master sys/dev/disk/nata ata-all.h ata-queue.c

Hasso Tepper hasso at estpak.ee
Tue Jan 20 00:39:02 PST 2009


Matthew Dillon wrote:
> This should improve the protection against large continuous linear
> reads and writes which can delay unrelated I/O's for very long periods
> of time.

This didn't improve the situation at least for me. I have following 
repeatable situation on my laptop (it's important to do it right after 
cold boot).

I start three terminal windows right after cold boot to X, top(1) in one 
of them and dd(1) in another:

dd if=/dev/zero of=./file bs=4096k count=2000

I watch at top(1) output until there is no free memory remaining (most of 
memory is shown as "inactive" now) and trying to execute ls in third 
terminal (listing my home directory). I takes about 20 seconds to listing 
appear. I press ^T ~ in every second and there is output:

load: 0.16  cmd: bash 1084 [biord] 0.00u 0.00s 0% 72k
load: 0.16  cmd: bash 1084 [spread] 0.00u 0.00s 0% 72k
load: 0.23  cmd: ls 1084 [spread] 0.00u 0.00s 0% 100k
load: 0.23  cmd: ls 1084 [spread] 0.00u 0.00s 0% 112k
load: 0.23  cmd: ls 1084 [spread] 0.00u 0.00s 0% 120k
load: 0.23  cmd: ls 1084 [spread] 0.00u 0.00s 0% 128k
load: 0.21  cmd: ls 1084 [spread] 0.00u 0.00s 0% 136k
load: 0.21  cmd: ls 1084 [spread] 0.00u 0.00s 0% 140k
load: 0.21  cmd: ls 1084 [spread] 0.00u 0.00s 0% 144k
load: 0.21  cmd: ls 1084 [spread] 0.00u 0.00s 0% 156k
load: 0.21  cmd: ls 1084 [spread] 0.00u 0.00s 0% 160k
load: 0.19  cmd: ls 1084 [spread] 0.00u 0.00s 0% 168k
load: 0.19  cmd: ls 1084 [spread] 0.00u 0.00s 0% 192k
load: 0.19  cmd: ls 1084 [spread] 0.00u 0.00s 0% 208k
load: 0.19  cmd: ls 1084 [spread] 0.00u 0.00s 0% 220k
load: 0.19  cmd: ls 1084 [spread] 0.00u 0.00s 0% 232k
load: 0.19  cmd: ls 1084 [spread] 0.00u 0.00s 0% 280k
load: 0.26  cmd: ls 1084 [spread] 0.00u 0.00s 0% 288k
load: 0.26  cmd: ls 1084 [spread] 0.00u 0.00s 0% 296k
<at this point the listing follows>
. ..

The laptop has 2GB of memory, no swap and 20GB Hitahci SATA harddisk.


-- 
Hasso Tepper





More information about the Commits mailing list