BIND - named crashes

Jim Reid jim at rfc1035.com
Wed Apr 12 23:28:44 UTC 2000


>>>>> "Ed" == Ed Zimmermann <edz at iswest.net> writes:

    Ed> I'm running BIND 8.2.2-P5 - Pent166 96meg on BSDI 4.0 One
    Ed> crashes every night the only comon element I found is in the
    Ed> log file: Apr 8 04:48:44 ns1 named[4761]: savedata: memget

    Ed> Do I need more RAM memory?  Hard disk space is over 60% avail.

The name server is running up to some OS-imposed limit on the size of
the process's data space. Consult the man pages for the limit command
- setrlimit() system call - on how to change this. Pay attention to
the defaults in the login classes in BSD/OS. Putting a clause like
	datasize unlimited;
in named.conf's options{} statement should do the trick. Beware that
in BSD/OS, the limit command tells lies, so you might need to use
sysctl to patch the kernel variable vm.maxdsize to something more
suitable.

IIRC, the default maximum data size for a process in BSD/OS is 64MB.
So if named is getting that big, you're getting uncomfortably close to
your computer's limit on physical RAM. So adding more memory would be
a good idea, but only if you can be sure the OS will let the name
server use it. (See above.) As a general rule, name server processes
should always be completely resident in RAM. They essentially access
all of their address space at random which makes life hard for the VM
subsystem's paging algorithms. And you don't want the name server to
get a page fault for any queries it gets because some page of its
cache has been written to a swap partition.





More information about the bind-users mailing list