Reposting in case any missed it the first time
On Wed, Dec 31, 2008 at 11:07 AM, Asif Iqbal <vad...@gmail.com> wrote: > Is there a patch for for version 2.0.4? > > Tried to apply the patch (both in Linux and Solaris using GNU patch) > from here http://kerneltrap.org/Linux/Finding_Bugs_With_CFS > > iqb...@ghar-iqbala:~/src/iperf-2.0.4$ patch -p1 < ../iperf-cpu-usage-fix-patch > patching file compat/Thread.c > Hunk #1 FAILED at 405. > 1 out of 1 hunk FAILED -- saving rejects to file compat/Thread.c.rej > patching file src/Reporter.c > Hunk #1 succeeded at 113 with fuzz 2 (offset 2 lines). > Hunk #2 succeeded at 341 (offset 2 lines). > Hunk #3 succeeded at 349 with fuzz 1 (offset 2 lines). > Hunk #4 succeeded at 568 (offset 14 lines). > Hunk #5 FAILED at 591. > 1 out of 5 hunks FAILED -- saving rejects to file src/Reporter.c.rej > patching file src/main.cpp > Hunk #2 succeeded at 143 (offset 1 line). > > > Linux: ubuntu 8.10 > > strace shows it is waiting > > # ./src/iperf -s -D -w 256k > Running Iperf Server as a daemon > The Iperf daemon process ID : 15812 > > # strace -p 15812 > Process 15812 attached - interrupt to quit > accept(3, {sa_family=AF_INET, sin_port=htons(35687), > sin_addr=inet_addr("127.0.0.1")}, [16]) = 0 > getsockname(0, {sa_family=AF_INET, sin_port=htons(5001), > sin_addr=inet_addr("127.0.0.1")}, [16]) = 0 > recv(0, "\0\0\0\0\0\0\0\1\0\0\23\211\0\0\0\0\0\0\0\0\377\377\374"..., > 24, 0) = 24 > mmap2(NULL, 8392704, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, > -1, 0) = 0xb6d22000 > mprotect(0xb6d22000, 4096, PROT_NONE) = 0 > clone(child_stack=0xb75224b4, > flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS|CLONE_PARENT_SETTID|CLONE_CHILD_CLEARTID, > parent_tidptr=0xb7522bd8, {entry_number:6, base_addr:0xb7522b90, > limit:1048575, seg_32bit:1, contents:0, read_exec_only:0, > limit_in_pages:1, seg_not_present:0, useable:1}, > child_tidptr=0xb7522bd8) = 15829 > accept(3, > > Solaris 10: > > # /usr/local/bin/iperf -s -D -w 256k > # pstack 29619 > 29619: /usr/local/bin/iperf -s -D -w 256k > ff0457b0 accept (3, 32c00, 32d00, 1) > 00015e2c _ZN8Listener3RunEv (30290, 2f990, 0, 2f800, ffffffff, 30890) + a4 > 00014e94 listener_spawn (2fd58, 0, ff073700, 0, feed0200, 1) + 60 > 0001befc thread_run_wrapper (2fd58, fee7c000, 0, 0, 1be20, 1) + dc > ff045238 _lwp_start (0, 0, 0, 0, 0, 0) > > (now run the iperf client to do a test and at completion it is now yeilding) > # pstack 29619 > 29619: /usr/local/bin/iperf -s -D -w 256k > ----------------- lwp# 2 / thread# 2 -------------------- > ff0457b0 accept (3, 33410, 33510, 1) > 00015e2c _ZN8Listener3RunEv (30290, 2f990, 0, 2f800, ffffffff, 30890) + a4 > 00014e94 listener_spawn (2fd58, 0, ff073700, 0, feed0200, 1) + 60 > 0001befc thread_run_wrapper (2fd58, fee7c000, 0, 0, 1be20, 1) + dc > ff045238 _lwp_start (0, 0, 0, 0, 0, 0) > ----------------- lwp# 4 / thread# 4 -------------------- > ff045350 lwp_yield (0, 0, ff073700, 0, feed0a00, 0) + 8 > 00017774 EndReport (36040, 32b88, 0, 0, 0, ff075380) + 28 > 00014edc server_spawn (32b88, 0, 0, 0, feed0a00, 1) + 18 > 0001bec4 thread_run_wrapper (32b88, fed7c000, 0, 0, 1be20, 1) + a4 > ff045238 _lwp_start (0, 0, 0, 0, 0, 0) > > Both Solaris and Linux shows CPU usage to 90% + even there is no iperf > test going on anymore > > On Solaris I did a dtrace probe and got similar data > > # dtrace -n 'syscall:::entry {...@[execname] = count();} tick-30s {exit(0);}' > dtrace: description 'syscall:::entry ' matched 232 probes > CPU ID FUNCTION:NAME > 1 51420 :tick-30s > [..] > iperf 8999610 > > ouch! 8999610 system calls in 30 secs even though there is no iperf > test is going on now. iperf gone mad!! > > Ok, lets find out what is iperf doing by tracing its pid > # dtrace -n 'syscall:::entry /pid == $target/ {...@[probefunc] = > count();} tick-30s {exit(0);}' -p 29995 > dtrace: description 'syscall:::entry ' matched 232 probes > CPU ID FUNCTION:NAME > 0 51420 :tick-30s > > yield 7519587 > > It yielded (??) 7519587 times in 30 secs.. wow!! > > Now probing the usermode stack to see what yeild is doing > > (root)@ftp0:~$ dtrace -n 'syscall::yield:entry /pid == $target/ > {...@[ustack()] = count();} tick-30s {exit(0);}' -p 29995 > dtrace: description 'syscall::yield:entry ' matched 2 probes > CPU ID FUNCTION:NAME > 0 51420 :tick-30s > > > libc.so.1`lwp_yield+0x4 > iperf`EndReport+0x28 > iperf`server_spawn+0x18 > iperf`thread_run_wrapper+0xa4 > libc.so.1`_lwp_start > 2931219 > > in 30 seconds it 2931219 times!! > > All these functions were touched by the patch to fix (not sure if that > fixed it) version 2.0.2 maybe?! > > -- > Asif Iqbal > PGP Key: 0xE62693C5 KeyServer: pgp.mit.edu > -- Asif Iqbal PGP Key: 0xE62693C5 KeyServer: pgp.mit.edu ------------------------------------------------------------------------------ Check out the new SourceForge.net Marketplace. It is the best place to buy or sell services for just about anything Open Source. http://p.sf.net/sfu/Xq1LFB _______________________________________________ Iperf-users mailing list Iperf-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/iperf-users
