Good point Wilko. I retract my observation. I didn't think of that. Andy On Fri, 20 Nov 2015, Wilko Kroeger wrote: > > Hello Andy > > I guess most of the stat/open calls are from opening shared libraries and > directories. Depending on your LD_LIBRARY_PATH there could be many > failed attempts to open or stat a file. > For example opening libc.so I see three attempts: > open("/opt/xrootd/prod/lib/libc.so.6", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) > open("/reg/g/psdm/sw/releases/dm-current/arch/x86_64-rhel7-gcc48-opt/lib/libc.so.6", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) > open("/lib64/libc.so.6", O_RDONLY|O_CLOEXEC) = 3 > > Also the strace didn't include the '-f' flag so not all threads are traced > and therefore the reads are very low. > > I ran "strace -c -f" for a 24 GB file and I see: > > % time seconds usecs/call calls errors syscall > ------ ----------- ----------- --------- --------- ---------------- > 71.19 50.881974 5098 9981 1097 futex > 17.94 12.818864 312655 41 1 nanosleep > 9.99 7.138469 23 314920 36189 read > 0.79 0.564963 14 40227 epoll_wait > 0.03 0.020346 14 1479 pwrite > 0.02 0.016862 6 2973 epoll_ctl > 0.02 0.011628 8 1494 sendto > 0.01 0.007300 456 16 munmap > 0.01 0.007009 467 15 brk > 0.00 0.000850 9 91 write > 0.00 0.000734 56 13 6 stat > 0.00 0.000572 11 50 26 open > 0.00 0.000434 7 66 mmap > 0.00 0.000239 6 37 close > 0.00 0.000156 4 43 mprotect > 0.00 0.000152 30 5 madvise > 0.00 0.000103 5 21 fstat > 0.00 0.000064 32 2 2 access > 0.00 0.000062 7 9 socket > 0.00 0.000059 15 4 tgkill > 0.00 0.000053 6 9 recvmsg > 0.00 0.000045 6 7 poll > 0.00 0.000045 45 1 execve > 0.00 0.000043 6 7 2 connect > 0.00 0.000017 6 3 getsockname > 0.00 0.000010 10 1 bind > 0.00 0.000010 2 6 fcntl > 0.00 0.000007 4 2 getsockopt > 0.00 0.000006 1 8 uname > 0.00 0.000004 2 2 getpeername > 0.00 0.000004 1 3 geteuid > 0.00 0.000004 4 1 arch_prctl > 0.00 0.000002 1 2 readv > 0.00 0.000001 1 1 getegid > 0.00 0.000000 0 1 lseek > 0.00 0.000000 0 2 rt_sigaction > 0.00 0.000000 0 1 rt_sigprocmask > 0.00 0.000000 0 6 clone > 0.00 0.000000 0 1 readlink > 0.00 0.000000 0 1 getrlimit > 0.00 0.000000 0 2 getuid > 0.00 0.000000 0 1 set_tid_address > 0.00 0.000000 0 2 2 openat > 0.00 0.000000 0 7 set_robust_list > 0.00 0.000000 0 1 epoll_create1 > 0.00 0.000000 0 1 pipe2 > ------ ----------- ----------- --------- --------- ---------------- > 100.00 71.471091 371566 37325 total > > > Cheers, > Wilko > > > On Fri, 20 Nov 2015, Andrew Hanushevsky wrote: > >> Yes, I also am mystified by some of the counts like: >> >> 0.01 0.001123 10 107 87 open >> 0.00 0.000000 0 16 14 stat >> >> among others. Why so many opens? Why are most of the opens and stat calls >> returning an error? It does seem consistent run to run. Quite strange. >> >> Andy >> >> On Fri, 20 Nov 2015, Lukasz Janyst wrote: >> >>> You could try to figure out which version of BSD supports socket watermarking and run a test with that. I strongly suspect that excessive returns from epoll are the cause here. Socket watermarking does not work on Linux. http://lkml.iu.edu/hypermail/linux/kernel/0412.1/0680.html >>> >>> --- >>> Reply to this email directly or view it on GitHub: >>> https://github.com/xrootd/xrootd/issues/20#issuecomment-158479186 >> >> >> --- >> Reply to this email directly or view it on GitHub: >> https://github.com/xrootd/xrootd/issues/20#issuecomment-158534692 > > > --- > Reply to this email directly or view it on GitHub: > https://github.com/xrootd/xrootd/issues/20#issuecomment-158559524 --- Reply to this email directly or view it on GitHub: https://github.com/xrootd/xrootd/issues/20#issuecomment-158562561 ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the XROOTD-DEV list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1