Timo Sirainen schrieb:
Is it actually the user CPU load or system CPU load? I'd think most of the time is spent on moving the new message to cur/ directory where there are the 58k messages.
Most time on System.
Cpu(s): 10.3%us, 87.1%sy, 0.0%ni, 1.7%id, 0.0%wa, 0.3%hi, 0.7%si,
0.0%st
Mem: 775148k total, 747620k used, 27528k free, 124464k buffers
Swap: 524280k total, 60k used, 524220k free, 186376k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+
COMMAND
10573 mailacce 19 0 20732 17m 5616 R 96.1 2.3 2:13.57
imap
10566 ingo 15 0 283m 273m 3544 S 1.3 36.2 3:33.49
imapsync
10602 root 15 0 2364 956 756 R 0.7 0.1 0:00.44 top
mail_nfs_storage: yes
Disable this during the sync, it'll add delays. And if you don't have more than one NFS client accessing the same mailbox at the same time, you don't need it anyway.
Also see if maildir_very_dirty_syncs=yes helps.
set both parameter, no speedup. Im now on message 48000 and it takes around 5 seconds and more to transfer one message.
strace output
strace -tt would have been better, it would have shown timestamps then too. But it didn't look like it did anything unexpected.
found attached a short strace with -tt ;-) is it normal, that he spent so much in entries like
08:48:30.167492 getdents64(7, /* 51 entries */, 4096) = 4080