On Apr 25, 2007, at 2:57 PM, Jeffrey Goldberg wrote:
> My contribution to the pool is easily able cope with the requests
> that its getting except that I am running up against a limit that
> appears to be built into the kernel handling of udp requests.
>
> My logs are routinely filled with things like
>
> Apr 25 11:36:21 n118 ntpd[56093]: too many recvbufs allocated (40)
> Apr 25 11:36:21 n118 last message repeated 3 times
It's not the kernel; that message comes from /usr/src/contrib/ntp/
libntp/recvbuff.c around lines 161 & 225:
/*
* Check to see if we're below the low water mark.
*/
if (free_recvbufs <= RECV_LOWAT)
{
if (total_recvbufs >= RECV_TOOMANY)
msyslog(LOG_ERR, "too many recvbufs
allocated (%ld)",
total_recvbufs);
[ ... ]
It appears that you can change this definition of RECV_TOOMANY here:
% grep -n RECV_TOOMANY */*
include/recvbuff.h:18:#define RECV_TOOMANY 40 /* this is
way too many buffers */
As far as I can tell, this only happens if ntpd gets a whole slew of
requests at once before it can generate some replies, and it simply
results in ntpd dropping the requests...
--
-Chuck
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
