Performance Issue for bind-9.4.2-P2 at Linux

[JINMEI Tatuya / 神明達哉]

At Wed, 17 Sep 2008 18:12:51 +0800,
"BroBind BroBind" <brobind at gmail.com> wrote:

> We upgraded our Caching-Publishing DNS servers from bind-9.2 to
> bind-9.4.2-P2  at our 4 IBM x346 2 Core CPU 2.80GHz  RHEL4-U3 x86_64 and
> 16GB Mem.  But we had experienced problems w/ it with too many open files
> problem. We resolved it by increasing the FD_SETSIZE at
> /usr/include/bits/typesizes.h and defining the
> STD_CDEFINES="-DISC_SOCKET_FDSETSIZE=4096" during compilation. Here's our
> configure options:
> 
> STD_CDEFINES="-DISC_SOCKET_FDSETSIZE=4096" ./configure
> --prefix=/var/bind/v9.4.2-P2
> --build=x86_64-redhat-linux-gnu --host=x86_64-redhat-linux-gnu
> --target=x86_64-redhat-linux-gnu \
> --with-libtool --enable-threads --enable-ipv6 --with-pic --with-openssl=yes
> --enable-largefile \
> --disable-openssl-version-check
> 
> Using iptraf, we're having total rates of around 20Mbps of traffic per
> server.  named.stats.txt showed that failure is very high and recursions
> compare w/ success. CPU utilization averages at 40-60 per server and memory
> rise to 20 to 50 % Utilization. We've seen many SERVFAIL from our tcpdumps
> going to legitimate domains and about 3-5 times dig to a domain before the
> queries became sucessful.

Please provide some more information:

- OS kernel version
- you named configuration (named.conf)
- initial log messages when you start named with the same command-line
  options plus '-g'
- number of sockets named actually opens when you see the many
  SERVFAILs (use lsof for example to see this)

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.