For the last several months we've been running r16375 on a bunch of
Netgear WGT634U's.  On #openwrt-devel, nbd mentioned that he had fixed
a memory leak in madwifi, so having noticed that our memory
(free+buffers+cache) seemed to decay over a period of about a week
(the rate varies) to the point where our captive portal (which forks
10 processes every auth) would tend to fail.

Aware of the reported fix, I was looking forward to rev'ing up to a
recent revision.  However, having flashed r18405 on an active node, I
am seeing what looks like even more rapid decay in available memory
than before.  We are running approximately the same user space apps as
before, the biggest memory consumers seem to be olsrd, perl+nocatauth
and openvpn.  I am taking occasional snapshots of /proc/$pid/statm,
and those three do seem to change in size a bit, with olsrd seeming to
grow a bit.  However, it doesn't seem enough to explain the ~2meg/day
decay in memory.

Here is an indicative graph:

  
http://silt.personaltelco.net:8080/cacti/graph.php?action=zoom&local_graph_id=253&rra_id=3&view_type=

Note that the last day (far right) has an even steeper slope of the
top yellow edge (representing the sum of free+buffers+cache) than the
rest of the graph (each reset of the roughly weekly sawtooth coincides
with a reset).

I am looking for help/suggestions in tracking down where the memory is
going.  Help?


-- 
Russell Senior         ``I have nine fingers; you have ten.''
[email protected]
_______________________________________________
openwrt-devel mailing list
[email protected]
https://lists.openwrt.org/mailman/listinfo/openwrt-devel

Reply via email to