Posted May 13, 2011 5:27 UTC (Fri) by bazsi (subscriber, #63084)
Parent article: Stable pages
I'm modifying a page in place at a high rate to keep the internal state of syslog-ng intact even in the case of a crash:
syslog-ng is following a logfile, its current position is kept in a file-backed memory region. In case the daemon is crashed the position remains there, so we can continue where we left off.
syslog-ng can update that page 100k/sec (especially if there are multiple such files and multiple threads reading), and I'm sure it's not waiting for writeback all the time, but that would probably negatively affect this and similar workloads.