On 08/ 9/16 01:46 AM, Robert Mustacchi wrote:
On 8/7/16 2:40 , Ian Collins wrote:
One thing that did surprise me testing on his box is the software builds
significantly faster in a KVM than in an LX zone. The difference is
most noticeable with linking where one target takes well over twice as
long in the zone.
KVM:
$ time ninja bin/Posix_Debug/Tests
[1/1] LINK bin/Posix_Debug/Tests
real 1m3.693s
user 0m55.470s
sys 0m7.772s
Zone:
$ time ninja bin/Posix_Debug/Tests
[1/1] LINK bin/Posix_Debug/Tests
real 2m45.004s
user 1m35.427s
sys 1m7.567s
Well, I'd probably start with two angles. The first is where do the
microstates say you're spending your time during the link. Let's make
sure there's no LAT on the scene. Otherwise, I might do something like
profile where we're spending our SYS time and try and break that down
further. Maybe just running the individual link command with ptime -m
would be a good start.
PID USERNAME USR SYS TRP TFL DFL LCK SLP LAT VCX ICX SCL SIG
PROCESS/LWPID
30176 1005 84 16 0.0 0.0 0.0 0.0 0.0 0.0 0 452 .1M 0
x86_64-linux/1
Lots of system calls.
Using dtrace for a couple of seconds:
# dtrace -n 'lx-syscall:::entry { @num[pid,execname] = count(); }'
dtrace: description 'lx-syscall:::entry ' matched 676 probes
^C
30090 x86_64-linux-gnu 4019849
Over the run time of the link (top few):
# dtrace -n 'lx-syscall:::entry { @num[probefunc] = count(); }'
dtrace: description 'lx-syscall:::entry ' matched 676 probes
^C
mmap 87
stat 334
close 1299
open 1407
fstat 1834
fcntl 2414
brk 15685
write 627589
read 3185104
lseek 13520417
Which is what I'd expect for a link that consumes 3.5GB of RAM...
--
Ian.
-------------------------------------------
smartos-discuss
Archives: https://www.listbox.com/member/archive/184463/=now
RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00
Modify Your Subscription:
https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb
Powered by Listbox: http://www.listbox.com
