Date: Wed, 1 Apr 2009 15:38:29 +0800 From: =?GB2312?B?tN7R0mNjdWl5eWFuQHNpbmEuY29t?= <ccuiyyan@gmail.com> To: freebsd-current@freebsd.org Subject: Performance Tools in FreeBSD Message-ID: <4451ccf20904010038p64ed2540mec3249f89bb7b8c2@mail.gmail.com>
next in thread | raw e-mail | index | archive | help
Dear all:
I wonder to know which kind of performance tools FreeBSD hackers use
to identify scalability or performance
bottlenecks. I really have a hard time in that. To identify
bottlenecks in our benchmarks, i try PMC and Dtrace.
However, PMC seems not to be supported in AMD Opteron CPU because i
get a "invalid arguments" error when
i use the command pmcstat -S instructions -O <outputfile>. However,
"instructions, cycles,..." are documented in
pmc manual. By the way, I enable the options in configuration file
options HWPMC_HOOKS and device hwpmc.
So i give up PMC and turn to Dtrace in BSD. Dtrace in FreeBSD can work,
but the information is meaningless.
Profile provider cannot be used. (When i used, the kernel goes into
kdb.) Tick provider can be used (It can only work
on one CPU by definition). However, when i use aggregation like
"dtrace -n 'tick-10ms /arg0!=0/{@ks[probefunc] = count();}'"
There are no human readable results. Only tools i can use is the
stack() command. However, the results are limited.
I know Dtrace is a tool in FreeBSD under development.
Are there any tools recommended to do kernel profiling or callgraph
effectively? (which functions take the most time)
Best Wishes!
Yan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4451ccf20904010038p64ed2540mec3249f89bb7b8c2>