Good day Al and all could you review couple patches that implement $subj for vfs and tmpfs. In short the idea is that we can protect operations taking semaphore related for set of names. definitely, protection at vfs layer isn't enough and filesystem will need to protect their own structures by itself, but in some cases vfs patch is enough. for example, tmpfs. during some loads one can see quite high load in /tmp. being mounted as tmpfs on big smp, we can get high contention on i_sem.
probably someone could try more-less real load? I wrote simple program that spawn few processes, then chdir to the given directory, then loops creating and unlinking file. The test box is dual PIII-1GHz: run 1: 2 processes create/unlink file on regular tmpfs [EMAIL PROTECTED] root]# mount -t tmpfs none /test [EMAIL PROTECTED] root]# (cd /test; time /root/crunl ./f 1000000 2) #1998: 1000000 iterations, create/unlink ./f-0-1998 #1999: 1000000 iterations, create/unlink ./f-1-1999 #384: done #384: done wait for completion ... OK real 0m36.224s user 0m0.823s sys 0m47.994s run 2: 2 processes create/unlink file on tmpfs + pdirops [EMAIL PROTECTED] root]# mount -t tmpfs -o pdirops none /test [EMAIL PROTECTED] root]# (cd /test; time /root/crunl ./f 1000000 2) #1992: 1000000 iterations, create/unlink ./f-0-1992 #1993: 1000000 iterations, create/unlink ./f-1-1993 #384: done #384: done wait for completion ... OK real 0m15.108s user 0m0.592s sys 0m29.406s run 3: 1 process creates/unlinks file on regular tmpfs [EMAIL PROTECTED] root]# mount -t tmpfs none /test [EMAIL PROTECTED] root]# (cd /test; time /root/crunl ./f 1000000 1) #2004: 1000000 iterations, create/unlink ./f-0-2004 #384: done wait for completion ... OK real 0m11.950s user 0m0.262s sys 0m7.465s run 4: 1 process creates/unlinks file on tmpf + pdirops [EMAIL PROTECTED] root]# mount -t tmpfs -o pdirops none /test [EMAIL PROTECTED] root]# (cd /test; time /root/crunl ./f 1000000 1) #2009: 1000000 iterations, create/unlink ./f-0-2009 #384: done wait for completion ... OK real 0m8.047s user 0m0.243s sys 0m7.646s 2 processes creating/unlinking on regular tmpfs cause ~200K context switches: procs memory swap io system cpu r b w swpd free buff cache si so bi bo in cs us sy id 2 0 0 0 1005760 6616 10928 0 0 0 0 1007 215095 1 64 35 2 0 0 0 1005760 6616 10928 0 0 0 0 1007 213580 1 67 32 2 0 0 0 1005760 6616 10928 0 0 0 0 1007 214445 1 63 36 2 0 0 0 1005760 6616 10928 0 0 0 0 1007 216250 1 63 36 2 processes creating/unlinking on tmpfs + pdirops cause ~44 context switches: procs memory swap io system cpu r b w swpd free buff cache si so bi bo in cs us sy id 2 0 1 0 1001824 6632 10912 0 0 0 0 1008 41 2 98 0 2 0 2 0 1002144 6632 10912 0 0 0 0 1008 45 2 98 0 2 0 2 0 1001632 6632 10912 0 0 0 0 1007 47 2 98 0 the next benchmark is rename. two processes generate random name, create file, generate new name, rename created file in new name and unlink: run 5: regular tmpfs [EMAIL PROTECTED] root]# mount -t tmpfs none /test [EMAIL PROTECTED] root]# (cd /test; time /root/rndrename ./f 1000000 2) #2036: 1000000 iterations #2037: 1000000 iterations wait for completion ... OK real 1m22.381s user 0m10.254s sys 1m50.214s run 6: tmpfs + pdirops [EMAIL PROTECTED] root]# mount -t tmpfs -o pdirops none /test [EMAIL PROTECTED] root]# (cd /test; time /root/rndrename ./f 1000000 2) #2044: 1000000 iterations #2045: 1000000 iterations wait for completion ... OK real 0m39.403s user 0m9.411s sys 1m8.626s thanks, Alex - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
