Andrei Dorofeev wrote:
So, unfortunately, the current situation with memory accounting
is not perfect *and* there's very little room for improvements.
The good news is that Memory Sets and Swap Sets projects
will solve two other problems you've identified :-)
The problem is essentially this - since many process can
share arbitrary pieces of memory - say by mapping a large
file MAP_PRIVATE and then writing to every other page -
the only accurate way of determining true sharing by
process is to walk all the pages in each process and figure
out which ones are sharing which pages... not very practical.
Given than some Solaris users have machines with .5 TB of RAM
and 20k processes, we need algorithms that are O(#processes)
rather than O(#processes * memory usage per process).
Overall, we can figure out pretty easily where pages are being
used, but attributing them to various processes is a lot of
work. Echo '::memstat' | mdb -k can be very informative:
# echo "::memstat" | mdb -k
Page Summary Pages MB %Tot
------------ ---------------- ---------------- ----
Kernel 140283 547 27%
Anon 95853 374 18%
Exec and libs 12203 47 2%
Page cache 161390 630 31%
Free (cachelist) 103333 403 20%
Free (freelist) 9064 35 2%
Total 522126 2039
Physical 522124 2039
- Bart
--
Bart Smaalders Solaris Kernel Performance
[EMAIL PROTECTED] http://blogs.sun.com/barts
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
