Performance tuning tips required for bind 9.6.1-P3!!!

[Shiva Raman]

Dear All

 This is in reference to the performance tuning , i had already gone through
the mailing list archives , but could not find answer to my
specific query mentioned here.

 I had installed  bind as a caching name  server for test purposes  and
planning to test performance that could give me around 10000 qps.

The os running Centos 5.4 64 bit , with minimal packages installed. The
server is dual quad core Intel Xeon 2.53 GHz ,16 GB RAM with 300 GB Hdd(
Raid 1) .

*Bind version installed is bind 9.6.1-P3 . Source extracted and compiled
with the following options. *

./configure --enable-epoll  --enable-atomic  --enable-ipv6 --enable-chroot
--with-openssl --with-randomdev=/chroot/named
--disable-openssl-version-check --with-libtool --enable-threads

Chrooted bind installation is done.

Only named, ssh and ntp services are running on the servers.

Right now i am using queryperf to test the performance with sample query
file of thousand entries. Right now
i am getting only 2000 to 2300 qps . I am writing querylogs to a separate
partition with noatime enabled for the
partition.

   OS hardening is done by  removing unwanted services, closing all
unneccesary ports and  securing the running services.

   My system is now using only 3 GB of RAM of total 16 GB.

*Following is the output of "uptime;free -m" during performance testing*

[root at localhost ~]# uptime;free -m
 22:19:52 up 1 day,  6:06,  3 users,  load average: 2.03, 2.06, 1.34
             total       used       free     shared    buffers     cached
Mem:         16047       3183      12864          0        238       2037
-/+ buffers/cache:        907      15140
Swap:         8189          0       8189




*Following is my named.conf*

acl testsetup_net {
            10.201.31.0/26; };

acl blacklistnets {
      192.0.2.0/24; 224.0.0.0/3; 10.0.0.0/8; 192.168.0.0/16;
};

// Main options defined here
options {
      directory "/conf";
      dump-file "named_dump.db";
      statistics-file "named.stats";
      pid-file "/var/run/named.pid";
      allow-recursion { localhost; ; testsetup_net; };
      allow-query { localhost; testsetup_net; };
      allow-query-cache { localhost;  testsetup_net; };
      allow-transfer { none; };
      blackhole { blacklistnets; };
      recursive-clients 20000;
      version "Not old!";
      datasize default;
      notify yes;
};

// Logging options are defined here.
logging {         // logging option for named  process
        channel "default_debug" {
            file "/logs/named.log" versions 10 size 50m;
           print-time yes;
            print-category yes;
             severity dynamic;
       };

      channel "queries" {         // logging option for queries to named
            file "/logs/query.log" versions 20 size 100m;
                print-time yes;
                print-category yes;
                severity dynamic;
        };

      category default { "default_debug"; };
     category queries { "queries"; };    // uncomment this to log queries
      category config { "default_debug"; };
      category security { "default_debug"; };
      category network { "default_debug"; };
      category lame-servers { null; };
        category edns-disabled { null; };
};


zone "." in {
      type hint;
      file "db.rootcache";
};

zone "localhost" in {
      type master;
      file "db.local";
      notify no;
};

zone "0.0.127.in-addr.arpa" in {
      type master;
      file "db.127.0.0";
      notify no;
};


   Kindly guide me for improving the bind performance from 2000 qps to
nearly 10000 qps. Which are the parameters i should change for improving
the performance? Any os level parameters to be changed for improving the
performance?

thanks in advance

Regards

Shiva Raman
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/bind-users/attachments/20100713/8f67e3aa/attachment.html>